Small Target Detection Algorithm Based on Receptive Field Amplification Feature Fusion

In order to solve the problems of high missed detection rate, low accuracy rate, and low recall rate in practical applications of small target detection, a small target detection algorithm based on receptive field amplification feature fusion is proposed. Firstly, add a receptive field amplification module to the basic network feature extraction which belongs to the full convolution single-stage target detection algorithm (FCOS) to improve the problem of less feature information extraction in the basic network ResNet-50 and low utilization of shallow feature layer information. Secondly, use the gating idea to filter the information fusion in the feature pyramid to reduce the interference of invalid information fusion. Finally, the attention mechanism module is added to the 7 feature layers to improve the accuracy of target positioning and classification. The experimental results on the COCO 2017 show that the detection accuracy of this algorithm is 2.4% higher than the traditional FCOS algorithm. Among them, the detection accuracy of small targets is increased by 3.2%, which gains better detection results. © 2023 Institute of Computing Technology. All rights reserved.