Improving the accuracy of pine wilt disease detection through the integration of multispectral imaging and machine learning

In this study, the accuracy of pine wood nematode detection was significantly improved by combining a UAV-mounted multispectral imaging system with machine learning algorithms, particularly the deep learning models SSD and YOLO v5. The study employed image preprocessing techniques, such as orthorectification and atmospheric correction, to ensure the accuracy and consistency of the data. On this basis, subtle differences in the health status of pine trees were revealed by analysing the spectral features extracted from the multispectral images. Traditional algorithms such as random forest and support vector machine, as well as deep learning models were used for classification and recognition, in which SSD and YOLO v5 demonstrated high accuracy in processing complex image data, reaching 0.92 and 0.93, respectively. This achievement not only improves the detection efficiency of pine wilt disease, but also provides a new technological pathway for forestry health management, which has an early diagnosis and treatment of the disease. important significance for early disease diagnosis and treatment.