Research on lightweight detection algorithm for millet quality grading.

Accurately assessing the quality of millet is of paramount importance for its storage, processing, and consumption. In complex environments, rapidly and precisely identifying densely packed millet poses a significant challenge in the quality grading process. This paper introduces an improved version of the YOLOv7 lightweight detection algorithm. The key enhancement involves integrating the super lightweight target detection algorithm ESNet. ESNet optimizes the lightweight design of the neck by enhancing the backbone network, thereby strengthening the network's feature extraction capabilities. Furthermore, by combining Soft non‐maximum suppression (NMS) with the EIOU Loss function, the accuracy of detecting dense millet in occluded scenarios is further optimized. Various models were evaluated using the millet dataset in this study. During the testing phase, images containing densely packed millet were randomly selected to assess the effectiveness of the models. The test results demonstrate that, compared to lightweight network architectures such as LCNet, ShuffleNetV2, and MobileNetV3, the improved YOLOv7‐ESNet model achieves higher average accuracy, precision, recall, and comprehensive evaluation metrics in detection. Additionally, by combining the YOLOv7‐ESNet model with Soft‐EIoU NMS, an average accuracy of 93.19% and an average prediction time of 52.5 ms were achieved. These results meet the real‐time grading requirements for millet quality, establishing a robust technical foundation for millet quality grading. This paper overcomes the challenge of precise millet detection in complex scenarios, providing a valuable contribution to the field. [ABSTRACT FROM AUTHOR]