CN-115205618-B - Earth surface coverage classification model training method, earth surface coverage classification method and earth surface coverage classification device

CN115205618BCN 115205618 BCN115205618 BCN 115205618BCN-115205618-B

Abstract

The invention discloses a surface coverage classification model training method, a surface coverage classification method and a device, wherein the surface coverage classification model training method comprises the steps of obtaining a training set; the method comprises the steps of inputting a sample image into an earth surface coverage classification initial model to conduct feature extraction to obtain a feature image, conducting first class feature processing on the feature image to obtain a first feature, predicting a first classification result by using the first feature, conducting second class feature processing on the feature image to obtain a second feature, extracting deep features of the first feature to obtain a third feature, fusing the second feature and the third feature, predicting a second classification result by using the fused features, calculating total loss of the first classification result and the second classification result by using labeling information, and conducting back propagation to adjust parameters of the earth surface coverage classification initial model. By the mode, the multi-level characteristics of the high-resolution remote sensing image can be fully utilized, and the accuracy of earth surface coverage classification is improved.

Inventors

Duan Fuzhi
WU CHENGLU
YU YANXUN
WANG YAYUN
Niu Zhongbin

Assignees

浙江大华技术股份有限公司

Dates

Publication Date: 20260512
Application Date: 20220531

Claims (13)

1. A method for training a surface coverage classification model, the method comprising: acquiring a training set, wherein the training set comprises a sample image and corresponding labeling information; inputting the sample image into an earth surface coverage classification initial model for feature extraction to obtain a feature image; Performing first-class feature processing on the feature image to obtain a first feature, and predicting a first classification result by using the first feature; Performing second class feature processing on the feature image to obtain a second feature, extracting deep features of the first feature to obtain a third feature, fusing the second feature and the third feature, and predicting a second classification result by utilizing the fused features; Calculating the total loss of the first classification result and the second classification result by using the labeling information and carrying out back propagation so as to adjust the parameters of the surface coverage classification initial model; The method comprises the steps of comparing a first classification result, a second classification result and corresponding labeling information respectively, calculating losses respectively, and carrying out back propagation by utilizing the total losses of the two classification results so as to adjust the parameters of the earth surface coverage classification initial model.
2. The method of claim 1, wherein the model training is performed, The first type of feature processing is first convolution processing, the second type of feature processing is second convolution processing, and the extracting deep features of the first features to obtain third features comprises: And processing the first feature through a pixel attention mechanism to obtain a third feature.
3. The method of claim 1, wherein the model training is performed, The fusing the second feature and the third feature, and predicting a second classification result by using the fused feature includes: And carrying out point multiplication on the second characteristic and the third characteristic, and predicting a second classification result by using a point multiplication result.
4. The method of claim 1, wherein the model training is performed, The labeling information comprises a small-class labeling and a large-class labeling, and the calculating the total loss of the first classification result and the second classification result by using the labeling information comprises the following steps: calculating a first loss by comparing the small category label with the first classification result, calculating a second loss by comparing the large category label with the second classification result, and calculating a weighted loss between the first classification result and the second classification result; And weighting and fusing the first loss, the second loss and the weighted loss to obtain the total loss.
5. The method for training a surface coverage classification model according to claim 4, it is characterized in that the method comprises the steps of, The calculating a weighted loss between the first classification result and the second classification result comprises: Calculating a third loss by comparing the small category labels with the corresponding second category results; and weighting and fusing the third loss and the second loss to obtain the weighted loss.
6. The method for training a surface coverage classification model according to claim 4, it is characterized in that the method comprises the steps of, The weighting fuses the first loss, the second loss and the weighted loss, and the obtaining the total loss includes: Setting weight values for the first loss, the second loss and the weighted loss, and performing regularization treatment on the weight values respectively; the total loss is a sum of the first loss, the second loss, a weighted value of the weighted loss, and a regularization result of the weighted value.
7. The surface coverage classification model training method of claim 6, wherein the total loss is: ; Where loss is the total loss, loss1 is the first loss, loss2 is the second loss, loss3 is the weighted loss, and w1, w2, w3 are weight values.
8. The method of claim 1, wherein inputting the sample image into an initial model of the earth surface coverage classification for feature extraction, obtaining a feature image comprises: extracting RGB three-channel images from the sample image to obtain an RGB image; Extracting features of the RGB images to obtain RGB feature images with multiple scales; Extracting features of the sample image, fusing the sample image with the RGB feature image with the same scale to obtain a downsampled feature image with the same scale, downsampling the downsampled feature image after fusion and fusing the downsampled feature image with the RGB feature image with the next scale to finally obtain a downsampled feature image with the small scale; And upsampling the small-scale downsampling characteristic image, fusing the small-scale downsampling characteristic image with the downsampling characteristic image of the same scale to obtain an upsampling characteristic image of the scale, upsampling the upsampling characteristic image after fusion and fusing the upsampling characteristic image with the downsampling characteristic image of the previous scale to obtain the characteristic image.
9. The method for training a surface coverage classification model according to claim 8, it is characterized in that the method comprises the steps of, The feature extraction of the RGB image to obtain RGB feature images with multiple scales includes: Extracting features of the RGB images to obtain four-scale RGB feature images; the feature extraction of the sample image comprises: and carrying out feature extraction on the sample image sequentially through a multi-head attention mechanism and a multi-layer perceptron.
10. The earth surface coverage classification model training method of claim 1, further comprising: acquiring a verification set, wherein the verification set comprises a verification image and a corresponding verification annotation image; Inputting the verification image into the surface coverage classification initial model, acquiring the precision of the surface coverage classification initial model by using the verification labeling image, and selecting the surface coverage classification initial model with the highest precision as a surface coverage classification model.
11. A method of surface coverage classification, the method comprising: Acquiring an image to be classified; Inputting the image to be classified into a surface coverage classification model to obtain a surface coverage classification result, wherein the surface coverage classification model is obtained by training a surface coverage classification promotion model by using the surface coverage classification model training method according to any one of claims 1-8.
12. A data device comprising a processor for executing instructions to implement the earth coverage classification model training method of any of claims 1-10 or the earth coverage classification method of claim 11.
13. A computer readable storage medium storing instructions/program data executable to implement the earth coverage classification model training method of any one of claims 1-10 or the earth coverage classification method of claim 11.

Description

Earth surface coverage classification model training method, earth surface coverage classification method and earth surface coverage classification device Technical Field The invention relates to the technical field of image processing, in particular to a ground surface coverage classification model training method, a ground surface coverage classification method and a ground surface coverage classification device. Background The formation and development of the earth's surface coverage type is not only affected by geographical factors, but is also closely related to human activity. The surface coverage classification is accurately carried out, so that on one hand, the distribution condition of natural resources in the area can be counted, on the other hand, the change condition of the natural resources can be counted along with the time, and the influence of human activities on the distribution of the natural resources is reflected and used for assisting in decision making. The traditional earth surface coverage type investigation adopts a field investigation mode, and huge manpower resource investment is needed. In recent years, with the development of remote sensing technology, the importance of earth surface coverage type investigation based on remote sensing images is increasingly highlighted, and earth surface coverage classification using high-resolution remote sensing images is one of important contents. The existing general earth surface coverage classification method for high-resolution remote sensing images mostly needs to be aimed at specific ground object artificial design characteristics, does not fully utilize the multiband characteristics of the high-resolution remote sensing images, and has certain limitations. Disclosure of Invention The invention mainly solves the technical problem of providing a ground surface coverage classification model training method, a ground surface coverage classification method and a ground surface coverage classification device, which can fully utilize the multi-level characteristics of high-resolution remote sensing images and improve the accuracy of ground surface coverage classification. The technical scheme includes that the earth surface coverage classification model training method comprises the steps of obtaining a training set, inputting a sample image into an earth surface coverage classification initial model to conduct feature extraction to obtain a feature image, conducting first type feature processing to the feature image to obtain a first feature, predicting a first classification result through the first feature, conducting second type feature processing to the feature image to obtain a second feature, extracting deep features of the first feature to obtain a third feature, fusing the second feature and the third feature, predicting the second classification result through the fused features, calculating total loss of the first classification result and the second classification result through the label information, and conducting back propagation to adjust parameters of the earth surface coverage classification initial model. The first type of feature processing is first convolution processing, the second type of feature processing is second convolution processing, and extracting deep features of the first features to obtain third features comprises the step of processing the first features through a pixel attention mechanism to obtain third features. The second classification result is predicted by utilizing the fused features, wherein the second classification result is predicted by utilizing the fused features by multiplying the second features by the third features and predicting the second classification result by utilizing the point multiplication result. The annotation information comprises a small-class annotation and a large-class annotation, and the step of calculating the total loss of the first classification result and the second classification result by using the annotation information comprises the steps of calculating the first loss by comparing the small-class annotation with the first classification result, calculating the second loss by comparing the large-class annotation with the second classification result, calculating the weighted loss between the first classification result and the second classification result, and carrying out weighted fusion on the first loss, the second loss and the weighted loss to obtain the total loss. The method comprises the steps of calculating a weighted loss between a first classification result and a second classification result, wherein the step of calculating a third loss comprises the steps of comparing a small class label with a corresponding second classification result, and the step of obtaining the weighted loss by means of weighted fusion of the third loss and the second loss. The method comprises the steps of weighting and fusing a first loss, a second loss and a weighted loss, wher