Lightweight Salix Cheilophila Recognition Method Based on Improved YOLOv8n

Abstract

Stumping is an important measure for the care and management of salix cheilophila during its growth. Rapid and accurate detection of salix cheilophila in the stumping period in desert is the basis of intelligent stumping equipment. However, the complex model needs high computing power of hardware. It limits the deployment and application of salix cheilophila recognition in intelligent stumping equipment. Therefore, this study took salix cheilophila in the desert areas of Shierliancheng, Inner Mongolia Autonomous Region in the stumping period as the research object, and proposed an improved YOLOv8 rapid identification method, named YOLOV8-VCAD. First, the lightweight network VanillaNet was used to replace the backbone of YOLOv8 to lessen the computing load and complexity of the model. Coordinate attention mechanism was embedded to extract important features by setting in location information, which strengthened the regression and positioning abilities of the model. Second, introducing an adaptive feature fusion pyramid network significantly strengthens the model's ability to characterize and integrate the features, improving the accuracy and performance of target detection. Finally, the CIoU loss in YOLOv8 is replaced by DIoU loss to quicken the regression convergence of the model. The experimental results show the accuracy of this method is 95.4%, and the floating-point a second (Flops) and parameters are 7.4G and 5.46M, respectively. Compared to the traditional YOLOv8, the precision of the proposed algorithm is increased by 7.7%, the recall is increased by 1.0%, the computational complexity is reduced by 16.8%, and the parameters are reduced by 7.9%. The performance of YOLOV8-VCAD for the detection of salix cheilophila in the stumping period is obviously better than the traditional YOLOv8. The algorithm proposed in this paper can quickly and accurately detect the salix cheilophila in the stumping period. Besides, it can reduce the deployment cost and difficulty of the vision module of intelligent stumping equipment, and provide technical support for the automatic intelligence of salix cheilophila stumping equipment.