Infrared Small Target Detection Algorithm Based on Improved DeepLabV3+

Infrared small target detection aims to segment small targets from the infrared images. Aiming at the problem of infrared small target detection, an improved algorithm is proposed based on DeepLabV3+ network. For the encoding part of DeepLabV3+, considering the small size of infrared small targets, the dilation rate of ASPP module is adjusted to capture the features of small targets more effectively. For the small targets are prone to lose spatial information during feature extraction, a location enhancement module is introduced to supplement accurate location information and to improve the accuracy of detection. In order to preserve detail information, an additional feature map is added from the backbone network to fuse detail information and semantic information for the decoding part of DeepLabV3+. Experimental results have demonstrated that the proposed algorithm is capable of enhancing the detection performance of DeepLabV3+ network in the infrared small target detection task.