AUTHOR=Sun Han , Wang Bingqing , Xue Jinlin 

TITLE=YOLO-P: An efficient method for pear fast detection in complex orchard picking environment

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 13 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2022.1089454

DOI=10.3389/fpls.2022.1089454

ISSN=1664-462X

ABSTRACT=Fruit detection is one of the key functions for automatic picking robot, but the accuracy is seriously reduced by the disordered background and other objects’ shade in orchard. Here, an effective mode based on YOLOv5, namely YOLO-P, was proposed to detect pears quickly and accurately. Shuffle block was used to replace the CBS structure of the second and third stages in YOLOv5 backbone, while the inverted shuffle block was designed to replace the fourth stage’s CBS structure. The new backbone could extract features of pears from long distance more efficiently. Convolutional block attention module (CBAM) was inserted into reconstructed backbone to improve the ability to capture the pear’s key feature. And Hard-Swish was used to replace the activation functions in other CBS structures of the whole YOLOv5 network. Then a weighted confidence loss function was designed to enhance the detection effect of small targets. At last, model comparison experiments, ablation experiments, daytime and nighttime pear detection experiments were carried out respectively. In the model comparison experiments, the detection effect of YOLO-P was better than other lightweight networks. The result showed that module’s average precision (AP) was 97.6% which was 1.8% higher than the original YOLOv5s. The model volume had been compressed by 39.4% from 13.7MB to only 8.3MB. Ablation experiments verified the effectiveness of proposed method. In the daytime and nighttime pear detection experiments, an embedded industrial computer was used to test the performance of YOLO-P under different background complexity and different shade degrees. The results showed that YOLO-P achieved the highest F1 score and FPS of 96.1% and 32, respectively. It was sufficient for picking robot to quickly and accurately detect pears in orchards. The proposed method can quickly and accurately detect pears under unstructured environment. YOLO-P provides support for the pear automated picking, as well as a reference for other types of fruit detection in similar environments.