Inversion of the forest dead fuel moisture content by UAV multisepectral image under the new leaves shade

Forest dead fuel moisture content (FDFMC) is an important factor affecting the occurrence and spread of forest fires. When the leaves have completely fallen, because of no leaves shade, the use of UAV multispectral cameras can achieve the spectral images easily. However, during the spring fire prevention period, it is difficult to obtain the full spectral images because of the shade of new leaves, therefore the inversion accuracy of FDFMC would be greatly affected by it. In this paper, an improved ConvNeXt convolutional neural network is proposed to predict FDFMC based on UAV multispectral camera data from 18 to 25 April 2025 in the urban forestry demonstration in Harbin City. A total of 6,031 sets of photos were captured using UAV multispectral camera, with each set containing six single-band images. The K-means clustering algorithm is used to segment the UAV multispectral images to extract the feature information for reducing the influence of new leaves shade. The trained model achieved 1.38% for MAE and 4.54% for RMSE. The experimental results showed that the improved ConvNeXt model can accurately predict the FDFMC. The new method proposed in this paper for predicting the FDFMC using the UAV multispectral images has feasibility and reference significance.