Highly overlapping strawberry leaf detection based on YOLOv5s-MBLS deep learning method

In response to the existing object detection algorithms are applied to complex strawberry planting scenes with poor detection accuracy, slow speed and difficult deployment, this paper proposes a lightweight YOLOv5s-MBLS detection algorithm. First of all, replace the CBS module in the YOLOv5 backbone network with the MBL module. The multi-scale convolution kernel of the MBL module can enhance the feature extraction ability of the backbone network while reducing the amounts of parameters and model size. Secondly, SIoU loss function is used to improve model accuracy and model detection speed. Finally, compared with the current popular target detection algorithm, the experimental results show that the the mean average precision (mAP) of YOLOv5s-MBLS algorithm is 1.6% higher than the original algorithm, the model size is reduced by 21.9%, and the average detection speed of each image is only 5.19 ms. Even compared with the latest YOLOv7, the mAP of YOLOv5s-MBLS is also improved by 0.4%, and other indicators are also better than YOLOv7, which proves the effectiveness of the algorithm.