Research on Strawberry Fruit Detection Based on YOLOv8

Aiming at the problems of strawberry fruit and stalk misdetection and missed detection due to leaf occlusion, aggregation, and stalk morphology in complex natural environments, and difficulty in detecting stalks from other green plants, an improved YOLOv8 strawberry fruit recognition and stalk detection algorithm model is proposed. The model incorporates the SA attention mechanism within its backbone, enhancing its capacity to prioritize essential features and object location data, thus improving feature extraction. Additionally, the SIoU loss function is implemented to emphasize target box size and shape details, expediting model convergence. A portion of the convolution within the model's detection head is substituted with the internally developed PRConv convolution, aiming to decrease redundant network parameters. Experimental findings indicate reductions in computational complexity and inference time by 9.9% and 7.4%, respectively, compared to the original model. Additionally, there's a notable improvement of 2.1% in [email protected]. Notably, this approach addresses the challenge of balancing detection time and accuracy, distinguishing it from other model experiments.