Improved Lightweight Apple Object Detection Method Based on YOLOv5s

Fast and accurate object detection is an important and challenging task in the context of automated apple harvesting. However, the current object detection models have relatively large network parameters. In this work, improvements have been made to the object detection algorithm based on YOLOv5s, by replacing the original model’s backbone network with the ShuffleNetV2 lightweight network structure. To compensate for the reduction in model parameters, this paper introduces the ECA mechanism and SPPF module after the ShuffleNetV2 modules to improve the model’s accuracy. Experimental results demonstrate that the improved lightweight YOLOv5s network reduces the parameter count by 85.6%, decreases computational load by 87.3%, and achieves a detection accuracy mAP@0.5 of 97.0% for apple detection, which is 0.4% higher than the original YOLOv5s model. The detection frame rate has also improved by 15 FPS compared to the original model, making it suitable for practical apple detection in real-world environments.