Road blocking information extraction based on deep learning image semantic segmentation

Abstract

The invention discloses a construction method of a road blocking image semantic segmentation sample library for full convolutional neural network training. The construction method comprises the steps of performing vectorization, enhancement and standardization on samples; secondly, introducing a classical convolutional neural network type and a network structure improvement method, and explaining a network realization method and a training process; then, using the full convolutional neural network obtained through training for conducting remote sensing image road surface semantic segmentation, and on the basis that the road surface which is not damaged after disaster is extracted, judging the road integrity through the length proportion of the road which is not damaged before disaster and after disaster. The precision evaluation indexes of the improved full convolutional neural network model are superior to those of an original full convolutional neural network model. The improved model is more suitable for specific problems of post-disaster undamaged pavement detection and road integrity judgment, and the adverse effects of tree and shadow shielding on road blocking information extraction can be effectively overcome.

Description

technical field [0001] The invention relates to the technical field of remote sensing monitoring. Specifically, it is the extraction of road blocking information based on deep learning image semantic segmentation. Background technique [0002] In the research of extracting road blockage information based on post-disaster single-temporal images, road vector data or other prior knowledge are often used as an aid. There are high requirements for registration accuracy between road vectors and images. In order to realize the automatic extraction of road blockage information under the condition of post-disaster emergency monitoring, it is necessary to focus on solving the impact on the accuracy of information extraction caused by the registration error between road vectors and images. Secondly, the image characteristics of the narrow road itself and linear distribution are not conducive to the application of traditional object-oriented image segmentation methods. It is difficul...

AI Technical Summary

Problems solved by technology

Secondly, the image characteristics of the narrow road itself and linear distribution are not conducive to the application of traditional object-oriented image segmentation methods

It is difficult to obtain a complete road surface during the segmentation process. When the road is shaded or covered by vegetation, it is easy to be misclassified as a blocked road, and it is also greatly affected by vegetation occlusion, which increases the difficulty of feature calculation and reduces the accuracy of classification.

Finally, the methods of most of the existing research results are relatively complicated, and it is difficult to ensure the dual requirements of accuracy and efficiency in the actual disaster emergency monitoring work

[0003] In particular, when constructing a general fully convolutional neural network, cross entropy (Cross Entropy) is often used as a loss function, but during the calculation of cross entropy, the weights of different types of pixels on the image are the same, which leads to When the unimproved fully convolutional neural network is directly applied to the problem of road semantic segmentation, a large number of correctly classified negative samples make the overall cross-entropy of the result low. When there are still many wrongly classified positive samples of the road, the network has already shown that the gradient disappears. state

That is, when the cross-entropy value is very low, the classification accuracy of the road positive samples is still not high and cannot be further improved

[0004] The main difficulty in obtaining point-by-point completeness detection results along the road from the semantic segmentation results of the undamaged road surface in the disaster area is that due to the existence of pre-disaster road vector compilation and image registration errors, the pre-disaster road vector and post-disaster remote sensing image It is difficult to achieve a complete match. Therefore, the pre-disaster road vector and the undamaged road surface extraction results from post-disaster remote sensing images cannot be completely matched, which makes it difficult to finally obtain the result of blocking road section discrimination or the point-by-point integrity detection results along the road.

Images