Research On Cultivated Land Extraction Technology Of Agricultural Remote Sensing Images Based On Deep Learning

Cultivated land is the basic guarantee for agricultural production.Accurately obtaining information on cultivated land area and distribution is of great significance for monitoring and production of agricultural resources.Remote sensing image classification is a very efficient method for extracting cultivated land information.For the classification of remote sensing images,a large number of algorithms have been proposed such as maximum likelihood classification,decision tree classification based on expert knowledge,and object-oriented classification methods.However,due to the different types of crops planted on cultivated land,the different time taken for remote sensing images,the different soil and soil properties of cultivated land,and so on,the existing algorithms do not have a good recognition effect on cultivated land.Therefore,at present the fine extraction of cultivated land still uses the most traditional artificial visual interpretation method,the efficiency is very low,and the accuracy of extraction is also restricted by the experience of personnel.Therefore,improving the efficiency and accuracy of arable land extraction in remote sensing images has become the key to determining whether remote sensing technology can play a real role in agriculture.Deep learning is a research field that has gradually emerged in recent years.At present,deep learning has become a mainstream research tool in the fields of speech recognition,image recognition,image classification,and target detection.Extracting arable land information from remote sensing images is exactly the problem of image recognition and classification.Therefore,using deep learning to extract arable land from agricultural remote sensing images is a very feasible research program.This paper is based on the theory of deep learning.After sorting the domestic and foreign research results in the field of remote sensing image classification and deep learning,the remote sensing image is preprocessed and annotated to make a training sample set.Finally,a convolutional neural network is used.CNN),as a classifier,carriecd out agricultural land extraction of remote sensing images and achieved the following main results:(1)The feasibility of deep learning in arable land extraction of agricultural remote sensing images was verified by consulting literature data and programming validation.It was concluded that deep learning is more suitable for cultureting agricultural remote sensing images than other traditional models.in conclusion.(2)A remote sensing image farmland extraction model was constructed using SVM model.After model trainng and adjustment,the final classification accuracy rate of the model was 73.39%,the recognition rate of fallow land was 83.58%,and the recognition rate of cultivated land was 79.33%.(2)Based on the Python language and tensorflow deep learning framework,a remote sensing image farmland extraction model based on convolutional neural network is programmed.The model’s overall classification accuracy rate reaches 82.82%,and the classification accuracy of fallow land and cultivated land is accurate.Reached 88.08%and 86.79%,respectively,and the overall classification accuracy of cultivated land features reached 87.85%.The classification result is better than the SVM model,and the classification noise in the image is less.The classification result image has a higher practical value.