首页 > , Vol. , Issue () : -
Objective：With large scale and high output value, China"s vegetable industry is a pillar industry to promote the increase of farmers" income and the development of rural agricultural economy. Rapidly and accurately obtaining vegetable crops planting structure information is of great significance for agricultural modernization, automation and precision. With the advantages of fast mobility, flexibility and image-spectrum merging, Unarmed Aerial Vehicle (UAV) hyperspectral remote sensing has wide prospects in crops fine classification. However, there are great variations in vegetable crop planting scales and modes and the fragmentation of agricultural landscape is high in China, and the vegetable crops are also affected by the coverage of plastic film, greenhouse and bird proof net, which easily produced the mixed spectral effect in UAV Hyperspectral images and also brings big challenges to the fine classification of vegetable crops. Method：Hyperspectral images of Gaoqiao scientific research base of Hunan Academy of Agricultural Sciences were obtained by UAV. According to the field survey, the area contains 14 ground feature categories, including eggplant, towel gourd, rice, pepper, tomato, and etc. Due to low requirements for data and excellent generalization ability, Support Vector Machine (SVM) is widely used in crops classification. And deep convolution neural network structure can automatically learn the abstract features of images and obtain higher-level and richer semantic information of samples, so as to better complete the classification task. For these reasons, SVM and Deep Learning (DL) methods were applied to the classification of vegetable crops in this study. Different from many hyperspectral classification verification experiments that randomly select training sets, training samples and test samples were manually selected in this study to reduce the spatial correlation between training sets and test sets. And the performance of different classification methods was evaluated using confusion matrix. Result：The results showed that using hyperspectral images obtained by UAV, the average overall accuracy of vegetable crops classification using SVM and DL methods is 78.03% and 90.75% respectively, and the average Kappa coefficients is 0.7359 and 0.8887 respectively. Compared with the SVM methods, the fine classification effects obtained by the DL methods are much more ideal, which is because the three-dimensional convolutional neural network and the convolutional neural network with attention mechanism can effectively extract the spectral spatial feature information in the image, thus shows a better performance in the classification of vegetable crops. The spatial texture characteristics of vegetable crops are obvious on large-scale plots, while they are various on small-scale plots, thus it is appropriate to use different DL methods for classification of vegetable crops on different scale plots. Conclusion：In this study, vegetable crops under different planting facilities were classified using UAV hyperspectral images. Under the influence of complex background such as plastic film, bird net and greenhouse, good performance was still achieved using SVM and Deep Learning (DL) methods, which can provide technology support for the modernization, automation and refinement of regional vegetable crop management.