A landslide detection method based on multi-scale feature fusion

Abstract

The invention discloses a landslide detection method based on multi-scale feature fusion, which relates to the technical field of target detection. Firstly, the captured images are sent to VGGNet withfocus module in real time. The depth features are extracted from the convolution layer from bottom to top, and then the depth features are fused from top to bottom according to different layers. Then, a selective search is performed on the original input image to generate several candidate boxes which may contain the target region. According to the size of the original image, the size of the candidate frame and the size of each feature map after feature fusion, the mapping size of each candidate frame on three feature maps is calculated. The dimensions of all the mapped feature maps are unified and passthrough all the connecting layers to adjust the position and dimensions of the candidate frames; Finally, the preliminary detection results are obtained by non-maximum suppression on the original image, and the final detection results are obtained. The invention can detect the debris flow phenomenon with different sizes, and greatly accelerates the detection speed.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a landslide detection method based on multi-scale feature fusion. Background technique [0002] my country's railway transportation is in the golden age of rapid development. The railway line is long, and a rail often passes through different terrains and rivers and lakes. Therefore, ensuring the safety of the railway line and its surrounding facilities has become the top priority. Every summer, there is a lot of precipitation, especially when there are sudden rainstorms and heavy rainstorms, landslides are extremely prone to occur in mountainous and hilly areas. Therefore, it is necessary to monitor the mountains and hills along the railway, and to report to the police the landslide phenomenon that occurs. [0003] The traditional method in the past is to use manual monitoring, but this method will greatly increase the cost of manpower and material resources, and it is...

AI Technical Summary

Problems solved by technology

[0007] The present invention aims at the above application background and technical difficulty, in order to solve the problem that landslides have various characteristics, landslides occupy a small size of the whole image, and landslides have inconsistent sizes; Aircraft aerial photography, use convolutional neural network to extract depth features, and use multi-scale depth feature fusion to perform multi-scale detection of landslides to enhance decision-making ability; specifically, a landslide detection method based on multi-scale feature fusion

Images