Performance Study of CBAM Attention Mechanism in Convolutional Neural Networks at Different Depths
Chunling Yang, Chunchao Zhang, Xuqiang Yang, Yanbin Li
Abstract
Compared with traditional imaging methods, infrared imaging has the advantages of strong anti-interference and good concealment, and is widely used in the fields of infrared alarm and reconnaissance. The target detection algorithm based on deep learning is much better than the traditional algorithm in target detection performance, but its small target detection performance is poor. The performance of small target detection can be improved by introducing attention mechanism. This paper aims to study the impact of adding Convolutional Block Attention Module (CBAM) to Convolutional Neural Network (CNN) on its performance, especially in small target detection. By adding CBAM at different depths, this paper explores the contribution of CBAM structure at different depths to the detection performance of small targets in convolutional neural networks. The results show that the small target detection performance of convolutional neural networks is significantly improved with CBAM attention mechanism, and the contribution of the performance on small target detection tasks varies by adding CBAM at different convolutional neural network structures depth.