Aiming at the problem that the accuracy of strawberry picking robot recognition of strawberry fruits in complex environments in the agricultural field is not high,a solution based on the improved YOLOv8 algorithm is proposed,which can realize accurate and fast recognition of strawberry fruits.Firstly,Mosaic data enhancement algorithm is utilized to preprocess target detection data,this method significantly improves the model's generalization ability and helps the model better identify strawberries in complex backgrounds.Secondly,the channel priority convolution attention mechanism is introduced.By focusing on information-rich channels in the image,this mechanism enhances the detection capability for small target strawberries,significantly improves feature extraction efficiency,and enables the model to learn and extract features related to strawberry recognition more intensively,thereby improves the precision of small target detection.Through a series of experimental verifications,the improved YOLOv8 algorithm performs significantly better than the original YOLOv8 algorithm in strawberry picking target detection,with a average precision mean of 89.35％,the average precision mean has increased by 5.83％compared with the original YOLOv8 algorithm.In summary,the proposed method has significant benefits in the identification of strawberry fruits,especially in dealing with small targets and complex backgrounds.The improved YOLOv8-EPCA network model has reached a level that can be applied in strawberry picking robots,it can provide strong support for real-time small target detection of picking robots in actual agricultural environments.
关键词
采摘机器人/YOLOv8算法/注意力机制/Mosaic数据增强算法
Key words
picking robot/YOLOv8 algorithm/attention mechanism/Mosaic data enhancement algorithm
维普
万方数据