Abstract: In view of the existing problems in strawberry recognition and detection, such as background interference, overlapping occlusion, and serious size difference, this paper proposes a detection method based on improved YOLOv5 algorithm. TC-SE module was introduced into YOLOv5's backbone and Neck, and feature fusion channel was added to PAN structure to improve small target detection performance. In the post-processing stage of model output, Gaussian weighting was introduced to improve the recall rate of overlapping occluded targets. The vector angle between regressions was introduced into the position loss function, and the penalty index was redefined to improve the recognition accuracy under complex background. The MAP of the improved model was 93%, which was 3.7 percentage points higher than the original model, and the detection speed was 31.15ms. The comparative experiment proves that the comprehensive performance index is better than other algorithms, and meets the real-time requirements of picking.