A Winter Jujube Defect Detection Method Based on the Improved Faster RCNN Model

Winter jujube sorting is an essential task before the jujubes enter the market. Effectively sorting and grading the harvested winter jujubes before selling can not only increase the income of jujube farmers but also enhance the brand recognition of winter jujubes. Research has found that the current winter jujube sorting work still relies on manual labor, and large-scale winter jujube sorting equipment, due to its high price, is not suitable for individual jujube farmers. Fortunately, with the development of deep learning, it has been widely applied in the field of agricultural product detection, making the intelligent and automatic sorting of winter jujubes possible. However, there is still a gap in the market for small-scale devices and algorithms suitable for small-scale winter jujube sorting equipment. Therefore, to address the issues of low accuracy, slow speed, and high cost of existing equipment in winter jujube sorting, this study proposes an improved Faster R-CNN model for small-scale winter jujube sorting equipment that identifies the external defects of winter jujubes. Firstly, the original VGG16 feature extraction network was replaced with ResNet50, and an SE module was added to ResNet50 to explicitly model the interdependencies between channels, allowing the network to adaptively recalibrate the feature responses of each channel. In this way, the network can focus more on useful features while suppressing irrelevant ones, thereby enhancing the representation capability of features. Secondly, the Feature Pyramid Network (FPN) was integrated to fully extract feature information at different scales. Finally, the original Non-Maximum Suppression (NMS) algorithm was replaced with the improved Soft-NMS algorithm to mitigate the issue of mistakenly deleting defect detection targets in the detected images, further improving the accuracy of winter jujube defect identification. The experimental results show that the improved Faster R-CNN model has a mean average precision (mAP) of 91.6% for winter jujube defect detection, with a detection speed of 32.5 frames per second. The mAP is 14 percentage points higher than SSD, 11.3 percentage points higher than YOLO v3, and 5.9 percentage points higher than YOLO v5. When the improved Faster R-CNN network model was deployed on a winter jujube detection platform to sort winter jujubes of different qualities, the recognition accuracies for high-quality fruit, damaged fruit, cracked fruit, insect-damaged fruit, and diseased fruit were 94.7%, 97%, 99.1%, 93.8%, and 99%, respectively. Although the detection time of two-stage object detection networks is longer than other networks, with the support of high-performance computers, the impact of detection speed can be neglected. The improved Faster R-CNN network model in this study can effectively and rapidly judge the appearance of winter jujubes, reducing the false detection rate in the winter jujube detection process. The research method can provide some ideas and theoretical support for workers related to the mechanization of the winter jujube industry.