P4CN-YOLOv5s: a passion fruit pests detection method based on lightweight-improved YOLOv5s

Passion fruit pests are characterized by their high species diversity, small physical size, and dense populations. Traditional algorithms often face challenges in achieving high detection accuracy and efficiency when addressing the complex task of detecting densely distributed small objects. To address this issue, this paper proposed an enhanced lightweight and efficient deep learning model, which is developed based on YOLOv5s, consists of the PLDIoU, four CBAM modules, and one newAnchors, termed P4CN-YOLOv5s, for detecting passion fruit pests. In P4CN-YOLOv5s, the Mosaic-9 and Mixup algorithms are initially used for data augmentation to augment the training dataset and enhance data complexity. Secondly, after analyzing the image set characteristics to be detected in this research, the point-line distance bounding box loss function is utilized to calculate the coordinate distance of the prediction box and target box, and aimed at improving detection speed. Subsequently, a convolutional block attention module (CBAM) and optimized anchor boxes are employed to reduce the false detection rate of the model. Finally, a dataset consisting of 6,000 images of passion fruit pests is used to evaluate the performance of the proposed model. The experimental data analysis reveals that the proposed P4CN-YOLOv5s model achieves superior performance, with an accuracy of 96.99%, an F1-score of 93.99%, and a mean detection time of 7.2 milliseconds. When compared to other widely used target detection models, including SSD, Faster R-CNN, YOLOv3, YOLOv4, YOLOv5, P4C-YOLOv5s, and YOLOv7 on the same dataset, the P4CN-YOLOv5s model demonstrates distinct advantages, such as a low false positive rate and high detection efficiency. Therefore, the proposed model proves to be more effective for detecting passion fruit pests in natural orchard environments.