Detection of Power Poles in Orchards Based on Improved Yolov5s Model

Abstract

During the operation of agricultural unmanned aerial vehicles (UAVs) in orchards, the presence of power poles and wires pose a serious threat to flight safety, and can even lead to crashes. Due to the difficulty of directly detecting wires, this research aimed to quickly and accurately detect wire poles, and proposed an improved Yolov5s deep learning object detection algorithm named Yolov5s-Pole. The algorithm enhances the model’s generalization ability and robustness by applying Mixup data augmentation technique, replaces the C3 module with the GhostBottleneck module to reduce the model’s parameters and computational complexity, and incorporates the Shuffle Attention (SA) module to improve its focus on small targets. The results show that when the improved Yolov5s-Pole model was used for detecting poles in orchards, its accuracy, recall, and mAP@50 were 0.803, 0.831, and 0.838 respectively, which increased by 0.5%, 10%, and 9.2% compared to the original Yolov5s model. Additionally, the weights, parameters, and GFLOPs of the Yolov5s-Pole model were 7.86 MB, 3,974,310, and 9, respectively. Compared to the original Yolov5s model, these represent compression rates of 42.2%, 43.4%, and 43.3%, respectively. The detection time for a single image using this model was 4.2 ms, and good robustness under different lighting conditions (dark, normal, and bright) was demonstrated. The model is suitable for deployment on agricultural UAVs’ onboard equipment, and is of great practical significance for ensuring the efficiency and flight safety of agricultural UAVs.

New Trends and Challenges in Precision and Digital Agriculture

“…In this regard, deep learning methods [29] such as convolutional neural networks (CNNs) are widely used. CNNs are used, among others, in the popular image detection algorithm-YOLO (You Only Look Once) [30][31][32][33]. In addition, CNNs can be successfully used to map various crops [34] or predict soil moisture in vegetated areas [35].…”

Defect Detection Algorithm for Battery Cell Casings Based on Dual-Coordinate Attention and Small Object Loss Feedback

“…Zabin et al [10] proposed a self-supervised representation learning model based on a contrastive learning framework, supported by an enhanced pipeline and a lightweight convolutional encoder, which effectively extracts meaningful representations from unlabeled image data and can be used for defect classification. For the battery casing, Zhang et al [11] proposed an improved YOLOv5s model, which can accurately and quickly detect three defects on the bottom surface of lithium batteries. YOLOv5s adds a layer to the network output layer to improve the detection effect of small defects, employs the convolutional block attention module attention mechanism to extract important information in the feature map, and then uses a new positional loss function to improve the position prediction accuracy of the model.…”