PDT: Uav Target Detection Dataset for Pests and Diseases Tree

TL;DR

This work addresses the lack of real-world UAV datasets for tree pest and disease detection and multi-class weed/crop classification. It introduces the PDT dataset for high-precision tree pest/disease detection in outdoor environments and the CWC dataset for detailed weed and crop classification, complemented by the YOLO-DP detector, which uses an Adaptive Large Scale Selective Kernel and GhostConv to handle dense, small targets. Experimental results across PDT, CWC, and SugarBeet2017 show YOLO-DP achieves state-of-the-art performance for dense UAV imagery and maintains strong classification capabilities, illustrating practical utility for precision agriculture. The authors also provide public access to the datasets and code, enabling broader research and potential deployment in real-world pest, disease, and weed management tasks.

Abstract

UAVs emerge as the optimal carriers for visual weed iden?tification and integrated pest and disease management in crops. How?ever, the absence of specialized datasets impedes the advancement of model development in this domain. To address this, we have developed the Pests and Diseases Tree dataset (PDT dataset). PDT dataset repre?sents the first high-precision UAV-based dataset for targeted detection of tree pests and diseases, which is collected in real-world operational environments and aims to fill the gap in available datasets for this field. Moreover, by aggregating public datasets and network data, we further introduced the Common Weed and Crop dataset (CWC dataset) to ad?dress the challenge of inadequate classification capabilities of test models within datasets for this field. Finally, we propose the YOLO-Dense Pest (YOLO-DP) model for high-precision object detection of weed, pest, and disease crop images. We re-evaluate the state-of-the-art detection models with our proposed PDT dataset and CWC dataset, showing the completeness of the dataset and the effectiveness of the YOLO-DP. The proposed PDT dataset, CWC dataset, and YOLO-DP model are pre?sented at https://github.com/RuiXing123/PDT_CWC_YOLO-DP.

Figures (11)

• Figure 1: Dataset comparison. (a) shows the PDT dataset (Low Resolution(LL) and High Resolution(LH)): 640×640, 5472×3648. (b) shows the characteristics of the CWC dataset: it contains 11 different similar plants. (c), (d) and (e) are the public datasets.
• Figure 2: Data example. (a) is a healthy goal and (b) is a unhealthy goal. The PDT dataset takes (b) as the category.
• Figure 3: PDT dataset generation and detection process. (a) represents the sliding window method, and (b) represents the "Human-in-the-loop" data annotation method. (c) means that a LL image is sent to the neural network for training, and LL and LH dual-resolution images are detected at the same time.
• Figure 4: YOLO-DP baseline model architecture. The FPNFPN+PANPAN module consists of GhostConvGhostConv, Upsample, Concat, and C3. C stands for Concat, S for Sigmoid, P for channel number dilation, $\times$ for matrix multiplication, and $+$ for matrix addition.
• Figure 5: Visualization of PDT dataset detection.