阅读说明：本技术 一种基于深度学习的圆形表盘识别方法 (Deep learning-based circular dial plate identification method ) 是由 杜辉 徐新兴 冯亮 董俊伟 章良杰 于 2019-12-12 设计创作，主要内容包括：发明提供一种基于深度学习的圆形表盘识别方法。该方法包括拍摄表盘图片、剔除不符合要求的图片、标记图像中的表盘、制作成数据集、用该数据集训练Yolo v3神经网络模型和测试训练模型等步骤。该方法解决了传统的表盘识别方法无法适用于复杂的场景下,没有通用性的缺点。(The invention provides a circular dial plate identification method based on deep learning. The method comprises the steps of shooting a dial picture, removing pictures which do not meet requirements, marking the dial in the image, making a data set, and training a Yolo v3 neural network model and a test training model by using the data set. The method overcomes the defects that the traditional dial plate identification method cannot be applied to complex scenes and has no universality.)

1. A circular dial plate identification method based on deep learning is characterized by comprising the following steps:

1) collecting a large number of dial plate image samples; the dial plate image sample comprises dial plates under various different visual angles and different scenes;

2) rejecting dial plate image samples which do not meet the requirements;

3) manually marking the dial plate in the dial plate image sample;

4) constructing a data set corresponding to the dial plate image sample;

5) inputting the data set into a Yolo V3 model for training and cross validation, and outputting a well-trained Yolo V3 model;

6) and acquiring a target dial image to be recognized, and inputting the image into a trained Yolo V3 model for recognition.

2. The deep learning-based circular dial identification method according to claim 1, wherein: and in the step 2), removing image samples with blurred images, low pixel quality and serious object occlusion.

3. The deep learning-based circular dial identification method according to claim 1 or 2, characterized in that: the step 3) is also provided with a relevant step of manually identifying the dial plate in the dial plate image sample.

4. The deep learning-based circular dial identification method according to claim 1 or 3, characterized in that: and 3), manually identifying the dial plate in the dial plate image sample. And (4) framing the dial plate in each image by using a rectangular frame, and marking the position and boundary information of the dial plate.

5. The deep learning-based circular dial identification method according to claim 1, wherein: in step 4, making the image label as an XML file; the XML file mainly comprises a size module, an object module and a path module; the size module mainly comprises pixel values and channel number information of the picture; the object module mainly comprises information about the name of the object and whether the position of the object appearing in the picture is specific or not; the path module calls a path for the picture.

6. The deep learning-based circular dial identification method according to claim 1, wherein: the Yolo V3 model contains 53 convolutional layers; a loss function in the training process adopts a cross entropy function; using softmax when predicting the object class, for an input image, YOLO V3 maps it to an output tensor of 3 scales, representing the probability of various objects existing at various positions of the image.

7. The deep learning-based circular dial identification method according to claim 1, wherein: and step 6) is followed by a correlation step of transmitting the recognition result of the Yolo V3 model to an output unit in a character and data stream mode, and the output unit outputs the recognition result.

8. The utility model provides a circular dial plate identification system based on degree of depth study which characterized in that: the device comprises an image acquisition unit, a neural network processing unit, an output unit and a storage unit; the image acquisition unit carries out manual annotation on the image sample and makes the image sample into a data set in an XML format; putting the data set into a neural network processing unit for training; after the training is finished, the output unit outputs the recognition result, and stores the picture in the jpg format into the storage unit.

Technical Field

The invention relates to the technical field of information, in particular to a circular dial plate identification method.

Background

The pointer instrument has the advantages of simple structure, low price, convenient use, electromagnetic interference resistance and the like, and is widely applied to various industries such as electric power, factories and mines, measurement and the like. In particular, in the power industry, pointer instruments are widely applied to monitoring the running state of power equipment, monitoring the gas pressure of a power switch and the like. With the pursuit of a power plant for quickly judging, positioning and stopping the loss of equipment operation faults, the conventional manual inspection, meter reading and recording modes cannot meet the requirements of the power plant on the management and production efficiency.

The general flow of the existing automatic reading technology of the meter reading is generally to acquire a meter image, image preprocessing, meter identification and meter reading. In the step of meter identification, identification of the image of the meter area is mainly identified. The currently more common methods are some optimization methods based on hough transform. However, the method has the defects of weak interference resistance, incapability of being applied to complex environments and weak universality.

Therefore, it is highly desirable to develop a method for identifying a circular dial with high versatility.

Disclosure of Invention

The invention aims to provide a circular dial plate identification method based on deep learning, and aims to solve the problems in the prior art.

The technical scheme adopted for achieving the purpose of the invention is that the circular dial plate recognition method based on deep learning comprises the following steps:

1) a large number of dial plate image samples are collected. Wherein the dial image samples comprise dials at a plurality of different viewing angles and under different scenes.

2) And rejecting the dial plate image samples which do not meet the requirements.

3) And manually labeling the dial plate in the dial plate image sample.

4) A data set corresponding to the dial image samples is constructed.

5) And inputting the data set into a Yolo V3 model for training and cross validation, and outputting a well-trained Yolo V3 model.

6) And acquiring a target dial image to be recognized, and inputting the image into a trained Yolo V3 model for recognition.

Further, in the step 2), image samples with unclear images, low pixel quality and serious object occlusion are removed.

Further, step 3) is also preceded by a related step of manually identifying the dial plate in the dial plate image sample.

Further, in the step 3), manual identification is carried out on the dial plate in the dial plate image sample. And (4) framing the dial plate in each image by using a rectangular frame, and marking the position and boundary information of the dial plate.

Further, in step 4, the image is marked as an XML file. The XML file mainly comprises a size module, an object module and a path module. The size module mainly comprises pixel values of pictures and channel number information. The object module mainly includes information on the name of the object and whether the position where the object appears in the picture is specific. The path module calls a path for the picture.

Further, the Yolo V3 model contained 53 convolutional layers. And a cross entropy function is adopted as a loss function in the training process. Using softmax when predicting the object class, for an input image, YOLO V3 maps it to an output tensor of 3 scales, representing the probability of various objects existing at various positions of the image.

Further, the step 6) is followed by a step of transmitting the recognition result of the Yolo V3 model to the output unit in a character and data stream manner, and the output unit outputs the recognition result.

The invention also discloses a deep learning-based circular dial plate recognition system which comprises an image acquisition unit, a neural network processing unit, an output unit and a storage unit. The image acquisition unit carries out manual annotation on the image sample and makes the image sample into a data set in an XML format. The data set is put into a neural network processing unit for training. After the training is finished, the output unit outputs the recognition result, and stores the picture in the jpg format into the storage unit.

The technical effects of the present invention are undoubted,

A. the dial plate at any position under a complex scene can be accurately identified, and the universality is very strong;

B. the precision is high, fast, reduces because of the mistake problem that manual meter reading caused.

Drawings

FIG. 1 is a flow chart of a method for identifying a circular dial;

FIG. 2 is a schematic diagram of a dial in a part of the collected scenes in example 1;

FIG. 3 is a schematic view of an unsatisfactory dial plate rejected in example 1;

FIG. 4 is a schematic view of a manual dial labeling;

FIG. 5 is a schematic diagram of an XML file;

FIG. 6 is a schematic diagram of a Yolov3 model of a neural network structure;

fig. 7 is a graph of the test results.

Detailed Description

The present invention is further illustrated by the following examples, but it should not be construed that the scope of the above-described subject matter is limited to the following examples. Various substitutions and alterations can be made without departing from the technical idea of the invention and the scope of the invention is covered by the present invention according to the common technical knowledge and the conventional means in the field.