Semantic Segmentation Of Agricultural Remote Sensing Image Based On Convolution Neural Network And Its Application

In recent years,As a large agricultural country,my country has a large scale of agricultural land.Effective use of agricultural land is of great significance to my country’s crop production and security.The image semantic segmentation method based on deep learning can effectively segment agricultural land and other categories,and monitor the use of agricultural land in multiple dimensions such as time and space in a long-term,real-time,and accurate manner.In recent years,UAVs,satellites and other equipment have invested heavily in the agricultural field to obtain a wealth of remote sensing image data resources of agricultural land,but manual interpretation has certain limitations.Therefore,this article proposes a method using convolutional neural networks.Carry out automatic interpretation to improve interpretation efficiency.The main contents of this article are as follows:1)Currently,there are too few public agricultural remote sensing image data sets,and the agricultural remote sensing image data set that this article needs to test is constructed.Use ArcGis software to separate and label the vegetation,buildings,water bodies,roads and other categories in the image and color them;then perform data enhancement operations on the image: set the size to 256×256 when randomly cropping,and set the cropped Small-size images are rotated by 90 degrees,180 degrees,and 270 degrees and flipped horizontally and vertically to expand the size of the data set;finally,the super-resolution reconstruction method of remote sensing images based on sub-pixel convolutional networks is used to improve the resolution of the image,and its prediction map The PSNR indicator is 22.6.2)Aiming at the poor accuracy of manual interpretation and early machine learning-based segmentation,this paper chooses to use deep learning-based semantic segmentation models,including Segnet network,DeeplabV3 network and DGCN network to train the constructed data set.After completing the training,use the accuracy evaluation indicators miou and accuracy to evaluate and compare the image semantic segmentation accuracy of different models.The miou value of the Segnet network is 0.59 and the accuracy value is 0.76;the miou value of the DeeplabV3 network is 0.61,and the accuracy value is 0.92;DGCN The miou value of the network is 0.17,and the accuracy value is 0.28;the final DeeplabV3 model has the best segmentation effect.3)Application of semantic segmentation of agricultural remote sensing images.The DeeplabV3 model is used as the core segmentation model of the application module,and then the agricultural remote sensing images around Hefei are collected and preprocessed,and then put into the module for segmentation.The results show that the "vegetation" category and the "water body" category can be accurately identified,Explains the practical application significance of monitoring agricultural land use.In summary,this article constructs a data set by manually labeling data.Through the comparison of the experimental results of different models,it is concluded that the use of the DeeplabV3 segmentation model can better achieve the semantic segmentation of agricultural remote sensing images.The result provides technical support for the segmentation of agricultural remote sensing images,and lays the foundation for the subsequent processing and analysis operations of the segmented categories,so it has practical significance.