阅读说明：本技术 基于人工智能的输液进度评估方法及系统 (Infusion progress evaluation method and system based on artificial intelligence ) 是由 陈贤勇 龙园 于 2021-09-22 设计创作，主要内容包括：本发明涉及人工智能技术领域,具体涉及一种基于人工智能的输液装置图像增强方法及系统。该方法包括：获取将吊瓶区域分为药液区域和无药液区域的模糊液面分界线,将模糊液面分界线的最小外接矩形作为目标图像；构建目标损失函数,目标损失函数包括图像增强目标损失函数、红色通道差量损失函数以及蓝色和绿色通道差量损失函数,根据目标损失函数得到最佳权重系数；采用最佳权重系数的多尺度视网膜增强算法得到清晰液面分界线；获取吊瓶的高度线段；根据清晰液面分界线和高度线段获取剩余药液的液位高度。有效的解决了患者在输液过程中,因为药液与吊瓶颜色接近导致视觉成像不明显,无法准确监测药液剩余量的问题。(The invention relates to the technical field of artificial intelligence, in particular to an image enhancement method and system of a transfusion device based on artificial intelligence. The method comprises the following steps: acquiring a fuzzy liquid level boundary dividing a transfusion bottle area into a liquid medicine area and a liquid medicine-free area, and taking a minimum external rectangle of the fuzzy liquid level boundary as a target image; constructing a target loss function, wherein the target loss function comprises an image enhancement target loss function, a red channel differential loss function and blue and green channel differential loss functions, and obtaining an optimal weight coefficient according to the target loss function; obtaining a clear liquid level boundary by adopting a multi-scale retina enhancement algorithm with an optimal weight coefficient; acquiring a height line segment of the infusion bottle; and acquiring the liquid level height of the residual liquid medicine according to the clear liquid level boundary and the height line segment. The problem of patient in infusion process, because the liquid medicine is close with the transfusion bottle colour and leads to the vision formation of image unobvious, can't accurate monitoring liquid medicine surplus has been effectually solved.)

1. An infusion progress evaluation method based on artificial intelligence is characterized by comprising the following steps:

acquiring a fuzzy liquid level boundary of a transfusion bottle area in an original image; the fuzzy liquid level boundary divides the infusion bottle area into a liquid medicine area containing liquid medicine and a liquid medicine-free area without liquid medicine;

selecting a minimum circumscribed rectangular area of the boundary of the fuzzy liquid level as a target image;

inputting the target image into a deep neural network, outputting an enhanced target image, fitting a multi-scale retina enhancement algorithm by the deep neural network by using a target loss function, wherein the target loss function comprises an image enhancement target loss function, a red channel differential loss function and blue and green channel differential loss functions, and obtaining an optimal scale as an optimal weight coefficient according to the target loss function; the difference of the liquid medicine area and the liquid medicine-free area in the red channel value is in a negative correlation with the red channel difference loss function; the difference loss function of the blue and the green channels is the sum of the difference of the blue channel values of the liquid medicine area and the liquid medicine-free area and the difference of the green channel values of the liquid medicine area and the liquid medicine-free area; when the quality of the enhanced target image is optimal, the optimal scale is the optimal weight coefficient, and a multi-scale retina enhancement algorithm with the optimal weight coefficient is adopted to enhance the target image to obtain a clear liquid level boundary;

acquiring a height line segment of the infusion bottle in the original image; and acquiring the liquid level height of the residual liquid medicine according to the clear liquid level boundary and the height line segment.

2. The method of claim 1, wherein the step of obtaining the blurred liquid level boundary of the region of the drop bottle in the original image comprises:

acquiring dark channel values of all pixels in the infusion bottle area according to a dark channel algorithm;

acquiring absolute values of difference values of the dark channel values and the gray values of all the pixels, preliminarily dividing the pixels into two types according to the absolute values of the difference values, and clustering different pixel classifications to obtain the liquid medicine areas and the liquid medicine-free areas;

and acquiring a boundary line between the liquid medicine area and the liquid medicine-free area as a fuzzy liquid level boundary line.

3. The method of claim 2, wherein the step of obtaining the dark channel values of all pixels in all of the original images according to a dark channel algorithm further comprises:

selecting the minimum value of each pixel in the original image in three RGB channels to form a gray scale map;

and taking each pixel as a central pixel in the gray-scale image, taking the central pixel as a central point of a window, and replacing the pixel value of the central pixel with the minimum value in the window so as to obtain the dark channel values of all pixel points in the original image.

4. The method of claim 1, wherein the step of constructing an image enhancement target loss function comprises:

and constructing the image enhancement target loss function by utilizing the fuzziness, wherein the fuzziness and the image enhancement target loss function have positive correlation.

5. The method of claim 1, wherein the difference in blue channel values and the difference in green channel values are positively correlated to the blue and green channel difference loss functions, respectively.

6. The method of claim 1, wherein the step of obtaining the height of the hanging bottle in the raw image comprises:

acquiring a plurality of connected domains in a transfusion bottle region in the original image, calculating the average length of the plurality of connected domains, selecting a target connected domain according to the average length, and taking a vertical line section formed by connecting two end points of the target connected domain as the height of the transfusion bottle.

7. The method of claim 1, wherein the step of obtaining the level height of the remaining medical fluid from the clear fluid level boundary and the height line segment comprises:

and acquiring the average coordinate of coordinates of all pixel points on the clear liquid level boundary line, wherein the distance between the average coordinate and the bottommost position of the height line segment is the liquid level height of the residual liquid medicine.

8. An artificial intelligence based infusion progress assessment system comprising a memory, a processor and a computer program stored in said memory and executable on said processor, characterized in that said processor implements said computer program when executing said computer program as claimed in claim 17 the steps of any one of the methods.

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to an infusion progress evaluation method and system based on artificial intelligence.

Background

The patient often needs to be transfused when receiving treatment, so that the patient or family members need to pay attention to the transfusion progress all the time to prevent the patient from being harmed due to untimely treatment when the liquid medicine is completely transfused. Although there are many ways of adaptively detecting the infusion progress by using a sensor, a corresponding detection device needs to be installed for each infusion device, which is not only unsuitable for popularization and application but also consumes manpower and material resources, and is easy to cause missed detection due to human negligence, thereby causing harm to patients.

Although the detection by using machine vision can overcome the defects of the above mode, the color of the liquid medicine of the patient is similar to that of the infusion device during infusion, which can cause the infusion surface to be not obvious enough in machine vision imaging, thereby affecting the final detection result.

Disclosure of Invention

In order to solve the technical problems, the invention aims to provide an infusion progress evaluation method and system based on artificial intelligence, and the adopted technical scheme is as follows:

in a first aspect, an embodiment of the present invention provides an infusion progress assessment method based on artificial intelligence, including the following steps:

acquiring a fuzzy liquid level boundary of a transfusion bottle area in an original image; the fuzzy liquid level boundary divides the infusion bottle area into a liquid medicine area containing liquid medicine and a liquid medicine-free area without liquid medicine;

selecting a minimum circumscribed rectangular area of the boundary of the fuzzy liquid level as a target image;

inputting the target image into a deep neural network, outputting an enhanced target image, fitting a multi-scale retina enhancement algorithm by the deep neural network by using a target loss function, wherein the target loss function comprises an image enhancement target loss function, a red channel differential loss function and blue and green channel differential loss functions, and obtaining an optimal scale as an optimal weight coefficient according to the target loss function; the difference of the liquid medicine area and the liquid medicine-free area in the red channel value is in a negative correlation with the red channel difference loss function; the difference loss function of the blue and the green channels is the sum of the difference of the blue channel values of the liquid medicine area and the liquid medicine-free area and the difference of the green channel values of the liquid medicine area and the liquid medicine-free area; when the quality of the enhanced target image is optimal, the optimal scale is the optimal weight coefficient, and a multi-scale retina enhancement algorithm with the optimal weight coefficient is adopted to enhance the target image to obtain a clear liquid level boundary;

acquiring a height line segment of the infusion bottle in the original image; and acquiring the liquid level height of the residual liquid medicine according to the clear liquid level boundary and the height line segment.

Preferably, the step of obtaining the fuzzy liquid level boundary of the transfusion bottle region in the original image comprises:

acquiring dark channel values of all pixels in the infusion bottle area according to a dark channel algorithm;

acquiring absolute values of difference values of the dark channel values and the gray values of all the pixels, preliminarily dividing the pixels into two types according to the absolute values of the difference values, and clustering different pixel classifications to obtain the liquid medicine areas and the liquid medicine-free areas;

and acquiring a boundary line between the liquid medicine area and the liquid medicine-free area as a fuzzy liquid level boundary line.

Preferably, the step of obtaining the dark channel values of all pixels in all the original images according to the dark channel algorithm further includes:

selecting the minimum value of each pixel in the original image in three RGB channels to form a gray scale map;

and taking each pixel as a central pixel in the gray-scale image, taking the central pixel as a central point of a window, and replacing the pixel value of the central pixel with the minimum value in the window so as to obtain the dark channel values of all pixel points in the original image.

Preferably, the step of constructing an image enhancement target loss function includes:

and constructing the image enhancement target loss function by utilizing the fuzziness, wherein the fuzziness and the image enhancement target loss function have positive correlation.

Preferably, the difference between the blue channel values and the difference between the green channel values are positively correlated to the blue and green channel difference loss functions, respectively.

Preferably, the step of obtaining the height of the infusion bottle in the original image comprises:

acquiring a plurality of connected domains in a transfusion bottle region in the original image, calculating the average length of the plurality of connected domains, selecting a target connected domain according to the average length, and taking a vertical line section formed by connecting two end points of the target connected domain as the height of the transfusion bottle.

In a second aspect, another embodiment of the present invention provides an artificial intelligence based infusion progress assessment system, comprising: memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the steps of the above method are implemented when the processor executes the computer program.

The embodiment of the invention has the beneficial effects that: the fuzzy liquid level boundary of liquid medicine in a transfusion bottle area is obtained through a dark channel algorithm, the transfusion bottle area is divided into a liquid medicine area and a liquid medicine-free area, self-adaptive selection is carried out on a Gaussian surrounding scale in a multi-scale retina enhancement algorithm through the difference value of each component in RGB three channels in the liquid medicine area and the liquid medicine-free area, the fuzzy liquid level boundary is processed through the obtained optimal scale, a clear liquid level boundary is obtained, then the residual amount of the liquid medicine is obtained through the height of the transfusion bottle and the average coordinate of pixel point coordinates on the clear liquid level boundary, and early warning is given through the residual amount of the liquid medicine. The problem of patient in infusion process, because the liquid medicine is close with the transfusion bottle colour and leads to the vision formation of image unobvious, can't accurate monitoring liquid medicine surplus has been effectually solved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart of a method for infusion progress assessment based on artificial intelligence according to an embodiment of the present invention.

Detailed Description

In order to further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the method and system for estimating the progress of infusion based on artificial intelligence according to the present invention with reference to the accompanying drawings and preferred embodiments shows the following detailed descriptions. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The embodiment of the invention is suitable for a hospital transfusion scene, the fuzzy liquid level boundary of liquid medicine in a transfusion bottle area is obtained through a dark channel algorithm, the transfusion bottle area is divided into a liquid medicine area and a liquid medicine-free area, self-adaptive selection is carried out on the Gaussian surrounding scale through the difference value of each component in RGB three channels in the liquid medicine area and the liquid medicine-free area, the optimal scale is obtained, the fuzzy liquid level boundary is processed, a clear liquid level boundary is obtained, the residual liquid medicine is obtained through the height of the transfusion bottle and the average coordinate of pixel point coordinates on the clear liquid level boundary, and early warning is given through the residual liquid medicine. The problem that the imaging is not obvious because the color of the liquid medicine is close to that of the infusion bottle is effectively solved, and the problem of timely judging the residual amount of the liquid medicine in the infusion process is solved.

The following describes a specific scheme of the infusion progress evaluation method and system based on artificial intelligence in detail with reference to the accompanying drawings.

Referring to fig. 1, a flowchart of a method for evaluating an infusion progress based on artificial intelligence according to an embodiment of the present invention is shown, in which the method includes:

step S100, acquiring a fuzzy liquid level boundary of a transfusion bottle area in an original image; the fuzzy liquid level boundary divides the infusion bottle region into a liquid medicine region containing the liquid medicine and a liquid medicine-free region without the liquid medicine.

First, an original image of a ward area is acquired.

Specifically, under the condition of normal and sufficient illumination, the RGB camera is placed above the ward to take a picture, the picture taking range of the RGB camera is required to cover all sickbed areas and corresponding infusion supports in the ward, and the image acquired by the RGB camera is an original image of the ward area.

And secondly, processing the acquired original image to acquire a transfusion bottle area. The infusion bottle is a container for containing liquid in a human body to be infused during infusion of a patient, such as an infusion bottle or an infusion bag, and the infusion bottle is hung on the infusion support to infuse the patient.

In the embodiment of the invention, a semantic segmentation network is adopted to process an original image, the semantic segmentation network adopts an encoder-decoder structure, and the specific training process is as follows:

1) the collected original image is used as a training data set, the training data set is labeled, a transfusion bottle in the transfusion device is labeled as 1, the other parts are labeled as 0, in the embodiment of the invention, 80% of the data set is randomly selected as the training set, and the other 20% of the data set is selected as a verification set.

2) Inputting original image data and label data into a semantic segmentation network, wherein an encoder is used for extracting image features and converting the number of channels into the number of labeled categories, and then a decoder is used for converting the height and width of an output image into the size of the input image so as to output the category of each pixel, wherein the pixel region labeled as 1 is the semantic segmentation image of the corresponding infusion bottle.

3) Training is performed using a cross entropy loss function.

And multiplying the obtained semantic segmentation image of the infusion bottle with the original image to obtain an infusion bottle area.

It should be noted that in the embodiment of the present invention, the evaluation is mainly performed on the infusion progress, and therefore, only the infusion bottle area in the original image is subjected to the subsequent analysis and calculation.

And finally, processing the transfusion bottle area by using a dark channel algorithm to obtain a fuzzy liquid level boundary.

It should be noted that the dark channel algorithm considers some pixels to have a very low value, even close to zero, in at least one color channel.

And selecting the minimum value of each pixel in the infusion bottle area in the RGB three channels to form a gray scale image, and then carrying out minimum value filtering on the pixel points of the gray scale image to obtain the dark channel value of each pixel point.

Specifically, each pixel is used as a central pixel in the gray-scale image, the central pixel is used as a central point of the window, and the minimum value in the window replaces the pixel value of the central pixel, so that the dark channel values of all pixel points in the transfusion bottle area are obtained.

It should be noted that, due to the scattering attenuation of light by the chemical solution and the selective absorption attenuation of light itself, there are differences between the scattering coefficient and the absorption coefficient of different wavelengths and the relationship between the scattering coefficient and the wavelength of light when the light propagates in normal air. Under the influence of the liquid medicine, the red light attenuation coefficient is far larger than the attenuation coefficients of the blue light and the green light, so that the gray value of the red dark channel is far smaller than the gray values of the blue dark channel and the green dark channel, and the image dark channel value of the transfusion bottle area finally containing the liquid medicine area tends to the gray value of the red dark channel.

Acquiring dark channel values of all pixels in the transfusion bottle area according to a dark channel algorithm; acquiring absolute values of difference values of all pixel dark channel values and gray values, preliminarily dividing pixels into two types according to the absolute values of the difference values, and clustering different pixel classifications to obtain a liquid medicine area and a liquid medicine-free area; a boundary line between the liquid chemical agent region and the liquid chemical free region is obtained as a blurred liquid surface boundary line.

Specifically, the dark channel algorithm is utilized to process the infusion bottle area, and the dark channel corresponding to each pixel point in the infusion bottle area is calculatedThe absolute value of the difference value between the channel value and the gray value of the RGB three channels is obtainedWherein, in the step (A),representing the absolute value of the difference between the dark channel value of the red channel and the RGB gray value,representing the absolute value of the difference between the dark channel value of the green channel and the RGB gray value,representing the absolute value of the difference between the dark channel value of the blue channel and the RGB gray value.

Because the obtained dark channel value is closest to the gray value of the red channel, when the absolute value of the difference value between the minimum dark channel value and the RGB gray value in the three channels of the pixel point is equal to the absolute value of the difference value between the dark channel value and the RGB gray value of the red channel, the pixel point is the pixel point containing the liquid medicine immediately. And processing each pixel point in the transfusion bottle area according to the dark channel algorithm, and classifying the pixel points containing the liquid medicine in the transfusion bottle area into one type, and classifying the pixel points not containing the liquid medicine into another type.

The method comprises the steps of preliminarily dividing the obtained pixel points containing the liquid medicine and the pixel points not containing the liquid medicine into liquid medicine areas, wherein a plurality of sub-areas possibly exist in the dividing process, and clustering the plurality of sub-areas.

Specifically, the embodiment of the invention adoptsThe average clustering algorithm clusters each subregion into two different regions, one is a region containing liquid medicine, and the other is a region without liquid medicine.

The two clustered regions are subjected to edge detection, and the method adoptsAnd the edge detection algorithm takes the overlapped part of the edge lines of the two areas as the boundary of the fuzzy liquid level.

Wherein the content of the first and second substances,the edge detection algorithm is a well-known technique, and will not be described in detail in the embodiments of the present invention.

And step S200, selecting the minimum circumscribed rectangular area of the boundary of the fuzzy liquid level as a target image.

Specifically, the distance between two pixels with the farthest abscissa distance is obtained as the length of a window, and the distance between two pixels with the farthest ordinate distance is obtained as the width of the window, so that a rectangular window area is obtained, the window area in the transfusion bottle area is extracted as a target image of the fuzzy boundary for subsequent analysis, subsequent calculation amount can be reduced for the target image analysis, and the influence of other areas on the accuracy of subsequent calculation results is avoided.

Step S300, constructing an image enhancement target loss function, a red channel differential loss function and blue and green channel differential loss functions to obtain an optimal scale as an optimal weight coefficient; the red channel difference loss function is the difference between the red channel value of the liquid medicine area and the red channel value of the liquid medicine-free area; the blue and green channel difference loss function is the difference between the blue channel value and the green channel value of the liquid medicine area and the blue channel value and the green channel value of the liquid medicine-free area; and enhancing the target image by adopting a multi-scale retina enhancement algorithm with the optimal weight coefficient to obtain a clear liquid level boundary.

The blurred liquid level boundary in the infusion bottle area is obtained in step S100, but due to noise interference and the influence of uneven illumination on the brightness distribution of the image, the obtained blurred liquid level boundary may not be a clear horizontal straight line, and in order to solve the problem that the blurred liquid level boundary is not clear, in the embodiment of the present invention, a multi-scale retina enhancement algorithm is adopted to process the target image, so as to eliminate the influence of illumination on the imaging of the blurred liquid level boundary.

The multi-scale retina enhancement algorithm can reduce or even eliminate the influence of incident light on the image, so that the aims of image color constancy, contrast enhancement and dynamic range compression can be well fulfilled. The algorithm mainly affects the parameters to be the Gaussian surrounding scale in the Gaussian surrounding function. When the gaussian surrounding scale in the gaussian surrounding function is selected too small, the contrast of the image can be well improved, but because the multi-scale retina enhancement algorithm assumes that the initial illumination image is changed slowly when the illumination image is estimated, namely the illumination image is smooth, but the illumination change of the image is not smooth at the edge of the area with larger brightness difference in the actual image, under the condition, the multi-scale retina enhancement algorithm can generate a halo effect when the image enhancement is performed on the area with larger brightness difference; when the gaussian surrounding scale in the gaussian surrounding function is selected too much, although the halo effect can be effectively avoided, the enhancement effect on details is not good enough, so that the boundary of the fuzzy liquid level is not clear enough to influence the judgment result.

When a multi-scale retina enhancement algorithm is adopted for image enhancement, the commonly used Gaussian surrounding scales are respectivelyAlthough the disadvantages of the single scale can be effectively reduced by using this method, the effect of image enhancement is relatively mediocre because the weight coefficients corresponding to the three gaussian surrounding scales are equal, and adaptive optimal weight distribution cannot be achieved in combination with specific image quality and application scenes, so that it is difficult to obtain the optimal image enhancement effect.

In the embodiment of the invention, a multi-scale retina enhancement algorithm is processed in a deep neural network training mode, and the weight coefficients of three Gaussian surrounding scales with the best enhancement effect when a target image is enhanced are determined, wherein the deep neural network adopts an encoder-decoder structure, the input is the target image, and the output is the image processed by the multi-scale retina enhancement algorithm.

The deep neural network training is as follows:

1) constructing an image enhancement objective function，In the embodiment of the invention, a secondary fuzzy algorithm is used for processing the target image by a multi-scale retina enhancement algorithm to obtain the ambiguity of the target image, wherein the ambiguity is a positive correlation between the ambiguity and an image enhancement target loss function.

It should be noted that the principle of the second fuzzy algorithm is as follows: if an image is blurred, when the image is subjected to blurring processing for the second time, the high-frequency component in the image is less in change; if the image is sharp, the high frequency components of the image may change very much when the image is blurred once. Therefore, the image definition can be determined according to the change of the high-frequency component, the higher the change of the high-frequency component is, the higher the image definition is, the lower the blurring degree is, after the blurring degree is normalized, the smaller the obtained result is, the clearer the corresponding image is, otherwise, the blurring degree is. By utilizing the secondary fuzzy algorithm, the definition evaluation can be carried out without a reference image, and the method is simpler and more efficient.

2) Constructing a Red channel delta loss function，Wherein, in the step (A),is the gray value of the red channel of the liquid medicine-free area,is the gray value of the red channel of the liquid medicine area,the difference between the red channels in the areas with and without the liquid medicine is in a negative correlation with a red channel difference loss function, which is intended to make the two types of areas in the output image as different as possible on the red channel.

3) Constructing blue and green channel delta loss functions，Wherein, in the step (A),is the gray value of the green channel of the liquid medicine-free area,is the gray value of the green channel of the liquid medicine area,is the gray value of the blue channel of the liquid medicine-free area,the gray value of the blue channel of the liquid medicine area, the difference of the blue channel and the difference of the green channel respectively form positive correlation with the loss function of the channel difference of the blue and the green, and the loss function aims to enable the two areas in the output image to be the same on the blue channel and the green channel as much as possible, so that the fuzzy liquid level boundary is clearer.

4) And updating the weight coefficients of the three Gaussian surrounding scales in the multi-scale retina enhancement algorithm by using a random gradient descent algorithm until the optimal scale with the best quality after image enhancement is obtained, and taking the optimal scale as the optimal weight coefficient.

In the embodiment of the invention, the multiscale retina enhancement algorithm adopts a Retinex image enhancement algorithm, and the secondary fuzzy algorithm adopts a Reblu algorithm.

And enhancing the target image according to the obtained optimal weight coefficient to obtain a clear liquid level boundary.

S400, acquiring the height of a transfusion bottle in an original image; acquiring the height of the residual liquid medicine according to the clear liquid level boundary; and obtaining the residual amount of the liquid medicine according to the height of the residual liquid medicine, and determining whether to give out an early warning or not according to the residual amount of the liquid medicine.

The method comprises the steps of obtaining a plurality of connected domains in a transfusion bottle area, calculating the average length of the connected domains, selecting a target connected domain according to the average length, and using a vertical line section formed by connecting two end points of the target connected domain as a height line section of the transfusion bottle.

Specifically, the edge contour of the infusion bottle area is processed by utilizing Hough line detection, a plurality of straight lines in the vertical direction are reserved, the straight lines are processed by utilizing a connected domain analysis method to obtain at least one connected domain, the average length of all the connected domains is calculated when a plurality of connected domains exist, the connected domain closest to the average length is obtained and is used as a target connected domain, two pixel points at two ends of the target connected domain are connected in a pixel coordinate system, the two pixel points have the same horizontal coordinate, a vertical line segment is obtained, and the vertical line segment is used as a height line segment of the infusion bottle.

And acquiring the average coordinate of coordinates of all pixel points on the clear liquid level boundary line, taking the distance between the average coordinate and the bottommost end of the height line segment of the infusion bottle as the height of the residual liquid medicine, and taking the height of the residual liquid medicine as the liquid level height index of the residual liquid medicine.

Specifically, in a pixel coordinate system, the distance between the boundary of the straight liquid level and the lowest end of the height line segment of the transfusion bottle is calculatedDifference between coordinate values of axesAnd when the height of the residual liquid medicine in the current infusion bottle is 0, namely the liquid level of the liquid medicine is 0, the liquid medicine is about to be infused, and early warning information is sent out to prompt medical staff to process in time. In other embodiments, the pre-warning liquid level height may be set as desired.

It should be noted that, in the embodiment of the present invention, the mouth of the default infusion bottle is oval, so that when the infusion bottle is used, the mouth is ovalThere is still a small amount of liquid medicine in the range of the oval bottle mouth, thereforeThe time forewarning has sufficient reaction time for medical staff to process.

In summary, in the embodiment of the present invention, the fuzzy liquid level boundary of the liquid medicine in the infusion bottle region is obtained through the dark channel algorithm, the infusion bottle region is divided into the liquid medicine region and the liquid medicine free region, the gaussian surround scale in the multi-scale retina enhancement algorithm is adaptively selected according to the difference between the components in the RGB three channels in the liquid medicine region and the liquid medicine free region, the optimal scale is obtained, the fuzzy liquid level boundary is processed, the clear liquid level boundary is obtained, the remaining amount of the liquid medicine is obtained further according to the height of the infusion bottle and the average coordinate of the pixel point coordinates on the clear liquid level boundary, and the warning is given according to the remaining amount of the liquid medicine. The problem of patient in infusion process, because the liquid medicine is close with the transfusion bottle colour and leads to the vision formation of image unobvious, can't accurate monitoring liquid medicine surplus has been effectually solved.

Based on the same inventive concept as the method embodiment, the embodiment of the invention also provides an infusion progress evaluation system based on artificial intelligence, which comprises: a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor, when executing the computer program, implements the steps of one of the above-described embodiments of the artificial intelligence based infusion progress assessment method, such as the steps shown in fig. 1. The infusion progress evaluation method based on artificial intelligence is described in detail in the above embodiments, and is not described again.

It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.