The Research Of Extracted Cannabis Information Based On High Resolution Remote Sensing Imagery

Cannabis is not just an industrial raw material but a drug plant,which contains tetrahydrocannabinol(THC)used for smoking..With the expansion of the scale of cultivation,China's cannabis production has been ranked first in the world.Due to the impact of China's planting patterns,cannabis' s cultivation distribute scattered,thus causing the country to lack effective and accurate means to collect cannabi's cultivation range and area information.Due to the wide variety of cannabis and the different planting conditions,we found that marijuana that is susceptible to biology mixed easily lead to changes in toxic content in the early investigation and monitoring work,so that part of cannabis toxic components(THC)content exceeded,therefore leave a serious risk for the proliferation of drugs.In this paper,the information extraction of cannabis plots is carried out by two different methods based on GF-2 satellite remote sensing images.Firstly,analyze the spectral features,spatial features and semantic features,and the corresponding eigenvalues were selected as decision tree nodes.The decision trees were extracted from the information extraction of cannabis plots.The original data were classified by ENVI software platform to realize the rapid identification and extraction of cannabis plots and finally get the desirable classification results.Secondly,by using the object-oriented classification method to extract the information of cannabis plots,the quadtree segmentation is used as the presegmentation of multi-scale segmentation to improve the segmentation efficiency.We try to combine the semantic information of normalized vegetation index NDVI as a single band Image segmentation.Finally,based on the single-band optimal scale calculation model,all the bands are jointly involved in the model calculation,and the optimal scale selection model is improved.The results show that:1.The results of the object-oriented information extraction method are greatly improved compared with the classification results of the pixel-based decision tree classifier,both in terms of user accuracy,producer accuracy and overall accuracy.2.The improved segmentation method makes the segmentation efficiency greatly improved,making a large area of cannabis cultivation monitoring possible.3.Using NDVI as a single band to participate in image segmentation also makes the segmentation accuracy improved,making the distinction between vegetation and non-vegetation more obvious.4.The optimal scale selection model takes into account the characteristics of all bands,and the optimal scale obtained from the model is more scientific and fundamentally ensures the accuracy of subsequent information extraction.