Abstract:

In this study, we use high-resolution remote sensing images to identify and extract the planting areas of wolfberry, a special economic crop in Qaidam, to provide evidence for market regulation and fine crop management. Taking Numuhong Farm, a typical wolfberry plantation area in the region as the research area, and using five classifiers of random forest, Softmax, support vector machine, BP neural network and maximum likelihood, we conducted the refined extraction of wolfberry plantation areas with different growth years, and verified the accuracy of the classification results. The results show that the classification effect using the random forest method is the best, with an overall classification accuracy of 93.8% and a Kappa of 0.93. Softmax, support vector and BP neural network methods have also achieved high classification accuracy, and their overall classification accuracy is 86.6%~87.6%, and the Kappa coefficient is 0.84~0.86. The maximum likelihood method has the worst classification effect, its overall classification accuracy is only 76.9%, and the Kappa coefficient is 0.73. Therefore, the use of domestic high-resolution satellites combined with a better classification method can realize the fine identification and monitoring of the planting areas and structures of minor characteristic cash crops such as Chinese wolfberry.

Key words: Qaidam, wolfberry, random forest, Softmax, SVM, BPNN