Significance target detection based on improved Pool Net
Can Cui, Zhe Ma · 2024
Firstly, this paper introduces the network structure of Pool Net, and improves on it by replacing the backbone network structure with the lightweight structure Mobilenetv3 to improve the operation speed; then introduces the CA module in the coordinate attention mechanism to enhance the feature information of the salient region; and introduces the loss function to reduce the information loss so that the salient region is more accurate; Finally verifies it by experimental indexes, which proves that the improved Pool Net model is capable of guarantee the accuracy of salient region and improve the speed of salient target detection.