Abnormal-Cut Tobacco Detection and Phenotypic Measurement Based on Improved YOLOv5s Rotating Frame

To accurately locate and count abnormal-cut tobacco in complex tobacco-making environments such as interference and occlusion, this study proposes an improved YOLOv5s method for detecting and counting abnormal-cut tobacco rotating frames. Firstly, the C3-DEBlock is constructed by combining the efficient multi-scale attention ( EMA ) module, dynamic snake convolution, and C3 module in the backbone network to adaptively adjust the receptive field, to enhance the feature extraction ability; Secondly, the context-anchor bidirectional feature pyramid network CAA-BiFPN is introduced into the neck network to capture the long-distance context information and improve the multi-scale feature fusion ability. Finally, the Kullback-Leibler divergence between Gaussian distributions is used as the regression loss function, so that the parameter gradient can be dynamically adjusted according to the characteristics of the object, to perform the regression of the detection box more accurately. Experiments show that compared with the mainstream object detection models FasterR-CNN, YOLOv4-tiny, and YOLOv5s, mAP increases by 14.91, 25.21, and 2.61 percentage points respectively. Regression analysis is performed by measuring the length and width of abnormal-cut tobacco manually, and the determination coefficients were 0.98, 0.985, 0.99, 0.985, 0.995, and 0.98, respectively. This method accurately locates abnormal-cut tobacco by rotating frame technology, which significantly reduces the interference of the background area. It provides an online detection scheme for accurate counting of abnormal-cut tobacco and precise grading control and optimization of cutting quality, which helps promote the modernization process of tobacco processing.