The automatic detection and identification of fish food processing factory images are of great significance for fishery products. However, due to the small size of fish food processing, target detection has considerable challenges and complex issues. In the last decade, numerous target detection methods for food have been proposed, such as methods based on infrared light, spatial-temporal joint processing models and human visual attention, but the detection of small food targets for food processing has not been fully investigated. In this regard, based on the characteristics of small fish food processing factories, a novel detection pattern based on improved You Only Look Once v5 (YOLOv5) is proposed to focus on the essential features of small fish food processing targets in this study. Compared with YOLOv5 anchor frames, YOLOv5 Small Target Detection (YOLOv5-STD) has an extra set of small selection boxes, which is sensitive to small food objects in fish processing factory images. By incorporating the optimization path aggregation network (PANet) function, the solution method of YOLOv5-STD is proposed, and its head neck architecture is optimized. A series of experimental results show that the proposed method can be used to detect small fish food processing targets more accurately and reliably than state-of-the art methods.