A Lightweight Method for Peanut Kernel Quality Detection Based on SEA-YOLOv5

Peanuts are susceptible to defects such as insect damage, breakage, germinant, and mildew, leading to varying qualities of peanuts. The disparity in peanut kernel quality results in significant differences in their prices and economic value. Conducting real-time, accurate, and non-destructive quality inspections of peanut kernels can effectively increase the utilization rate and commercial value of peanuts. Manual inspections are inefficient and subjective, while photoelectric sorting is costly and less precise. Therefore, this study proposes a peanut kernel quality detection algorithm based on an enhanced YOLO v5 model. Compared to other models, this model is practical, highly accurate, lightweight, and easy to integrate. Initially, YOLO v5s was chosen as the foundational training model through comparison. Subsequently, the original backbone network was replaced with a lightweight ShuffleNet v2 network to improve the model’s ability to differentiate features among various types of peanut kernels and reduce the parameters. The ECA (Efficient Channel Attention) mechanism was introduced into the C3 module to enhance feature extraction capabilities, thereby improving average accuracy. The CIoU loss function was replaced with the alpha-IoU loss function to boost detection accuracy. The experimental results indicated that the improved model, SEA-YOLOv5, achieved an accuracy of 98.8% with a parameter count of 0.47 M and an average detection time of 11.2 ms per image. When compared to other detection models, there was an improvement in accuracy, demonstrating the effectiveness of the proposed peanut kernel quality detection model. Furthermore, this model is suitable for deployment on resource-limited embedded devices such as mobile terminals, enabling real-time and precise detection of peanut kernel quality.