Research On Extraction Of Ginseng Planting Area Based On Remote Sensing Image

Ginseng is known as the "King of All Herbs" and has high medicinal and economic value,many people destroy the forest land to grow ginseng for their own benefit,which seriously damages the ecological environment,therefore,it is urgent to verify and control the deforestation of ginseng.The use of remote sensing images to extract ginseng cultivation areas can obtain the distribution of ginseng cultivation areas,providing powerful support for the verification and control of deforestation and ginseng cultivation.In recent years,with the development of remote sensing technology and deep learning,ginseng planting areas can be extracted from remote sensing images by using deep learning technology,but due to the limitations of deep learning technology and the imaging characteristics of remote sensing images,the extraction of ginseng planting areas from remote sensing images has the problems of complex background interference,large differences in the morphology of ginseng planting areas and difficulties in extracting small target ginseng planting areas.In view of the above problems,the work done in this paper includes the following three aspects:(1)Creating a ginseng cultivation area extraction dataset.In this paper,by manually screening the multi-source remote sensing image data of Heilongjiang Province,some of the dense ginseng cultivation areas were selected as sample data.Based on the imaging characteristics of ginseng cultivation areas in remote sensing images,the sample annotation tool was used to draw ginseng cultivation area labels,and after cropping by the program,the sample data was expanded by using data enhancement techniques to finally obtain the ginseng cultivation area extraction dataset.(2)A ginseng cultivation region extraction method based on the U-Net structure is proposed.To address the problem of complex background interference in ginseng cultivation region extraction,it is proposed to introduce a channel spatial attention module into the U-Net structure,so that the method can suppress complex background interference and focus more on ginseng cultivation region extraction.To address the problem of large differences in the morphology of ginseng cultivation regions,a convolutional layer is proposed to be added to each decoding layer of the U-Net to improve the method’s ability to understand the semantics of remote sensing images and to improve the method’s ability to extract ginseng cultivation regions of different morphologies.(3)A Deep Lab v2-based ginseng cultivation region extraction method is proposed.To address the problem of difficulty in extracting small target ginseng cultivation regions,it is proposed to use multi-scale convolution in the Deep Lab v2 coding structure to improve the method’s ability to extract ginseng cultivation regions at different scales,thus enabling better extraction of small target ginseng cultivation regions.By conducting experiments on the ginseng cultivation area extraction dataset and the pre-processed GID dataset,the results obtained show that the method proposed in this paper performs well on both datasets,effectively overcomes the problems of ginseng cultivation area extraction,improves the accuracy and recall rate of ginseng cultivation area extraction,and illustrates the effectiveness,generalization and robustness of the method in this paper.