Abstract:
Chinese herbs are mostly grown in mountainous areas where natural conditions are suitable, and their cultivation areas are complex, with irregularly fragmented plots, making it difficult to accurately calculate cultivation areas and estimate yields using traditional survey methods. Convolutional Neural Network (CNN) has a good effect in accurately and efficiently extracting the spatial distribution of crops in remote sensing images. Therefore, the feature extraction capability of CNN was combined with the application of Chinese herbal medicine, and the real-time monitoring of the planting area of
panax notoginseng, a Chinese herbal medicine resource in Wenshan Prefecture, Yunnan Province, was achieved by combining Sentinel-2 images with survey data. The overall accuracy of its extraction was 95.57% and the Kappa coefficient was 0.91, and there was confusion between some fine shadows and greenhouse of
panax notoginseng, but the overall extraction effect is better. The results show that
Panax notoginseng cultivation in Wenshan is concentrated in the north-western part of the prefecture, while the south-eastern part is less cultivated due to such comprehensive factors as topography and temperature. In addition, 93.57% of
panax notoginseng is planted on slopes below 25° due to its special requirements for soil moisture, while slope orientation is a factor affecting the length of sunlight and light intensity of
panax notoginseng, and 70.40% of
panax notoginseng is chosen to be planted on semi-shady and semi-sunny slopes.