Research on Improving YOLOv5 Small Target Detection Algorithm Based on 3D Artificial Intelligence Technology

Aiming at dim targets in images, the YOLOv5 image segmentation algorithm is presented in this paper. The effects of some parameters in the replacement model on the detection efficiency and inference speed were investigated. The output stage based on series neural network is added to realize the detection of small targets. Secondly, the scene extraction model CFM and the FSM model based on channel and spatial feature refinement are proposed to solve the semantic loss problem in multi-level feature fusion. In this way, the fusion problem of multi-scale features is solved. Its reasoning speed is 3 milliseconds longer than the original YOLOv5.