Intelligent electric power inspection tracking method based on target sensing depth feature compression

Abstract

The invention discloses an electric power intelligent inspection tracking method based on target perception depth feature compression, and the method comprises the steps: inputting a to-be-tracked electric power equipment picture into a pre-trained convolutional neural network, and extracting depth features containing to-be-tracked electric power equipment and a background region thereof; inputting the Gaussian soft label and a given Gaussian soft label into a single-layer convolutional network, and learning by using negative balance loss until the network converges; quantizing the back propagation gradient of each channel by using global average pooling to obtain an initial importance weight, and binarizing to obtain a final importance weight; during tracking, the depth feature of a target template is extracted only in the first frame, and the depth feature of a search area image in the next frame is extracted and compressed; and performing related operation on the compressed target template features and the search area features to obtain the position of the target. According to the method, the importance weight is obtained through the back propagation network gradient to guide the depth feature compression process, the concise depth features are obtained, the operation cost is reduced, and the tracking robustness is improved.

Description

technical field [0001] The invention relates to a target tracking method for deep feature compression, in particular to an electric power intelligent patrol tracking method based on target perception deep feature compression. Background technique [0002] A stable and reliable power system is the key to ensuring people's livelihood and economic development. In recent years, the application of vision-based detection technology in power detection has received extensive attention. Both robot-based detection technology and UAV-based detection technology need to track the determined target, so robust tracking methods are very important. The development of deep learning has led to the improvement of tracking performance. However, due to the deep feature parameters provided by the deep neural network model, the calculation and storage costs are high, it is difficult to meet the real-time requirements of intelligent detection, and it is difficult to deploy it on mobile intelligent ...

AI Technical Summary

Problems solved by technology

The development of deep learning has led to the improvement of tracking performance, but due to the deep feature parameters provided by the deep neural network model, the calculation and storage costs are high, it is difficult to meet the real-time requirements of intelligent detection, and it is difficult to deploy on mobile intelligent detection equipment

Images