阅读说明：本技术 一种基于时序遥感数据和卷积神经网络的作物种植分布预测方法 (Crop planting distribution prediction method based on time sequence remote sensing data and convolutional neural network ) 是由 张炜 黄河 史杨 吴晓伟 于 2019-10-11 设计创作，主要内容包括：一种基于时序遥感数据和卷积神经网络的作物种植分布预测方法,包括下列步骤：步骤1：地面调查及训练样本建立；步骤2：构造基于时序遥感数据和卷积神经网络的作物种植分布预测模型,所述卷积神经网络通过在多时相图像中目标像素点及其周围像素点的数据对其进行预测,输入值为多时相的高分辨率多光谱图像,输出值为作物类型、轮作方式的分类信息；步骤3：将统计区域的时序遥感数据输入已构建的模型获取识别结果。只需要通过少量的代表性地块的地面调查,构建了融合遥感遥感数据时序特征和遥感图像局部特征的预测模型,引入了决策点的上下文信息,提高了预测结果的准确性。(A crop planting distribution prediction method based on time sequence remote sensing data and a convolutional neural network comprises the following steps: step 1: ground investigation and training sample establishment; step 2: constructing a crop planting distribution prediction model based on time sequence remote sensing data and a convolutional neural network, wherein the convolutional neural network predicts a target pixel point and peripheral pixel points in a multi-temporal image through data of the target pixel point and the peripheral pixel points, the input value is a multi-temporal high-resolution multi-spectral image, and the output value is classification information of crop types and crop rotation modes; and step 3: and inputting the time sequence remote sensing data of the statistical area into the constructed model to obtain a recognition result. A prediction model fusing the time sequence characteristics of the remote sensing data and the local characteristics of the remote sensing image is constructed only through ground investigation of a small number of representative plots, context information of decision points is introduced, and accuracy of prediction results is improved.)

A crop planting distribution prediction method based on time sequence remote sensing data and a convolutional neural network is characterized by comprising the following steps: comprises the following steps:

step 1: ground investigation and training sample establishment;

step 2: constructing a crop planting distribution prediction model based on time sequence remote sensing data and a convolutional neural network, wherein the convolutional neural network predicts a target pixel point and peripheral pixel points in a multi-temporal image through data of the target pixel point and the peripheral pixel points, the input value is a multi-temporal high-resolution multi-spectral image, and the output value is classification information of crop types and crop rotation modes;

and step 3: and inputting the time sequence remote sensing data of the statistical area into the constructed model to obtain a recognition result.

The crop planting distribution prediction method based on time series remote sensing data and convolutional neural network as claimed in claim 1, characterized in that: the step 1 comprises the following steps:

step 1.1: researching and counting main planting crop types, crop rotation modes and key phenological period time nodes in the area;

step 1.2: and (3) aiming at different crop types and crop rotation modes in the step 1.1, selecting partial representative plots, and recording the positions and ranges of the plots, the planting areas, the planted crops, the planting time and the like in detail to obtain training samples.

The crop planting distribution prediction method based on the time series remote sensing data and the convolutional neural network as claimed in claim 2, characterized in that: the step 2 comprises the following steps:

step 2.1: collecting remote sensing images containing the representative land blocks selected in the step 1.2 at different periods, accurately registering the remote sensing images at different times, and carrying out pixel-by-pixel classification marking on areas with different crop types and crop rotation modes;

step 2.2: the time sequence remote sensing data is used as the input of the model, the classification information of the crop type and the crop rotation mode is used as the output of the model, and a multilayer convolution neural network is constructed;

step 2.3: and (3) fully training the parameters in the multilayer convolutional neural network by using the training samples in the step 2.1 and using a back propagation algorithm.

The crop planting distribution prediction method based on time series remote sensing data and convolutional neural network as claimed in claim 3, characterized in that: in the step 2, in the image processing process, the data of each waveband of the remote sensing images of different time phases are stored in the form of a plurality of color channel images, and when prediction is performed, the data of corresponding pixel points are obtained according to the field size, the time phase number and the channel number of the image of each time phase, so that prediction is performed on target pixel points.

The crop planting distribution prediction method based on the time series remote sensing data and the convolutional neural network as claimed in claim 4, characterized in that: the step 3 comprises the following steps:

step 3.1: collecting remote sensing images with the same time as that of the remote sensing images collected during neural network training in the research area range, and accurately registering the remote sensing images at different times;

step 3.2: and (3) inputting the registered time sequence remote sensing data in the step (3.1) into the multilayer convolutional neural network model trained in the step (2.3) to obtain a crop planting classification prediction result of the research area.

The method for predicting the crop planting distribution based on the time-series remote sensing data and the convolutional neural network as claimed in any one of claims 3 to 5, wherein: during neural network training, the time for acquiring the remote sensing image is from March to November every middle of the month.