The XVIII International Conference on Mine Surveying

20 / 2023-03-27 08:45:47

Pine Wood Nematode Disease Area Identification Based on Multi-Temporal Multi-Source Remote Sensing Images and an Attention U-Net Model

pine wood nematode disease area identification; remote sensing images; deep learning; transfer mechanism

Abstract: Pine wood nematode disease is a dangerous forest pest that can quickly kill trees, posing a serious threat to the health of forests. Identifying areas affected by this disease can help manage forest trees, remediate affected areas, and ensure forest ecological security. With the rapid development of remote sensing technology, remote sensing images provide advantages such as large area observation, high timeliness, and high resolution, making them ideal for accurately and quickly identifying pine wood nematode disease areas.      This paper proposes a deep learning approach to identify pine wood nematode disease areas using multi-temproal Muiti-Source Remote Sensing Images. Specifically, a pine wood nematode disease are identification attention U-Net(PWIAU-Net) model is developed and trained using multi-temporal Gaofen-3 and Beijing-2 images acquired at different stages of pine wood nematode disease. To verify the generalization ability of the proposed model, a transfer learning mechanism is also proposed to identify the best-trained model for different regions affected by pine nematode disease. Experimental results demonstrate that the proposed method significantly improves classification accuracy, with a 15-18% F1-score improvement compared to traditional methods. Moreover, the model exhibits pleasing generalization ability across different forest regions. Overall, this research provides a promising tool for identifying and managing pine wood nematode disease areas using remote sensing images and deep learning methods.