Research of an Enhanced YOLOv5-Based Algorithm for Garbage Detection in Service Areas

Addressing the challenge of effectively detecting service area litter through monitoring systems, this study introduces an improved YOLOv5 object detection algorithm. To begin, the backbone feature extraction network of YOLOv5 is replaced with the more powerful ResNet50, to enhance feature extraction capabilities. In addition to detect minute garbage more effectively, a dedicated small object detection layer is added, along with a corresponding feature extraction and fusion module. Furthermore, by removing the original large and medium object detection heads, the waste of computational resources is reduced, thereby accelerating the model's inference speed. Overall, NWD is introduced in place of the IoU loss function, further enhancing the detection accuracy for small objects. Experimental results demonstrate that the improved model excels in the task of service area litter detection, achieving an mAP0.5 of 91.95% and a 5.02% increase compared to the original YOLOv5s model, with a detection rate of up to 72 frames per second. Compared to other mainstream object detection models, the method of this study shows superior performance in both detection accuracy and inference speed.