赣南脐橙未成熟果实与叶片的高精度检测与识别OA

In the early stage of navel orange planting,accurate detection of fruits and leaves is of great significance for timely de-tection of pests and diseases,and improvement of fruit quality and yield.However,the peel and leaf color of immature navel or-anges are similar,and the actual planting scenarios are complex,so it is difficult for existing algorithms to balance accuracy and speed.To this end,based on the YOLOv8n model,the CPH-YOLOv8n algorithm is proposed,which replaces the traditional Concat module with a contextual attention splicing module with dynamic attention mechanism,effectively improving detection ac-curacy.In order to solve the problem of slow bounding box regression caused by the similar color of fruit peel and leaf,the Powerful-IoUv2 loss function is introduced to optimize the regression process and accelerate the convergence of the model.In or-der to realize the lightweight of the model,the wavelet transform downsampling module is introduced to reduce the number of model parameters and the computational complexity on the basis of ensuring the accuracy.Experimental results show that the im-proved algorithm performs well,and when maintaining a high accuracy of 95.3％of mAP0.5,the recall is significantly improved,the computational amount is reduced to 7.8 GFLOPs,and the detection time is reduced to 2 ms.Compared with the original algo-rithm,the recall rate is increased by 3.7 percentage points,the computational amount is reduced by 3.7％,and the detection speed is increased by 42.9％under the premise of comparable accuracy.It can be seen that the CPH-YOLOv8n algorithm can balance the accuracy and speed of fruit and leaf detection in complex environments.