Information Extraction Of Agricultural Greenhouses Basedon GF-6

Satellite remote sensing technology is widely used in many aspects of agricultural production because of its fast,non-destructive and large-scale characteristics in obtaining information.Combined with remote sensing technology,we can obtain agricultural land use,investigate agricultural resources,monitor crop growth,and estimate crop yield.In the process of developing and promoting modern agriculture,the first task is to quickly collect crop growth status,and remote sensing undoubtedly has obvious advantages in the rapid collection of crop information.GF-6 satellite belongs to the development project of gaofen-6 satellite in China.Its spatial resolution can reach 2 meters,and it has a good recognition of ground objects.Based on the characteristics of GF-6 satellite image,this paper selects the object-oriented classification method to extract agricultural greenhouse information.In object-oriented classification,the accuracy of classification is directly determined by whether the object fits the reality,so image segmentation is an important part.In the process of image processing,we often encounter the selection of segmentation method and segmentation scale.The optimal segmentation effect can be obtained by comparing the segmentation parameters of different scales.In order to select the appropriate segmentation scale,this paper use KNN(K Nearest Neighbor)classification method to classify the images under different segmentation scales,and obtains the optimal segmentation scale by comparing the classification accuracy.Before classification,55 image features commonly used in high score image classification are selected to construct feature space and optimize feature space.Then,combined with the optimal segmentation scale,cart decision tree classification,SVM(Support Vector Machine)classification and KNN classification are used to classify and compare the images,and the most suitable classification algorithm for greenhouse extraction is obtained,which is used to extract agricultural greenhouse information in Pingxiang City.Finally,according to the extraction results,the area and distribution of agricultural greenhouses in Pingxiang City were analyzed.The main work and achievements are as follows.(1)This paper introduces chessboard segmentation method,quadtree segmentation method,FNEA multi-scale segmentation method and spectral difference segmentation method.By comparing the segmentation effect and objective evaluation index,it is found that FNEA segmentation method has good applicability in high-resolution image segmentation.Combined with ESP tool to predict the optimal segmentation scale,in the {50,120} scale range,8 groups of segmentation scales are selected with 10 steps,and an experimental area is selected.The object-oriented KNN classification method is used to classify the experimental area images under different segmentation scales.In the {0.3,0.8} interval,taking 0.1 as the step size,the segmentation effect of agricultural greenhouse under different weight combination of shape factor and compactness factor was compared.Finally,the optimal segmentation scale and parameters suitable for agricultural greenhouse extraction research were obtained;(2)Selecting reasonable feature dimension and feature value can improve the accuracy and efficiency of classification.Combined with the spatial distribution features,texture features,geometric and spectral features of agricultural greenhouses,55 features commonly used in high score image classification are selected to construct feature space.By analyzing the relationship between the feature dimension and the sample separability,it is concluded that when the feature dimension is 26,there is a better separation between the samples and the classification effect is better.According to the importance order of the features after feature analysis,the first 26 image features are selected for subsequent classification and extraction;(3)Using the same segmentation parameters and features,CART decision tree method,SVM method and KNN method are used to classify the same experimental area respectively,and the classification effect and accuracy are compared to obtain the most suitable classification method for agricultural greenhouse information extraction.It is found that cart decision tree is better than the other two in overall classification accuracy and greenhouse extraction accuracy,and it is more suitable for agricultural greenhouse extraction;(4)According to the extraction results of agricultural greenhouses in Pingxiang City,the number of agricultural greenhouses in Pingxiang City is 1249,with a total area of3910947.82m2,and the development scale is small.Through the analysis of the distribution of greenhouses,it is concluded that greenhouses are mainly distributed in the areas close to roads and water flow,followed by the middle of farmland,and there are few greenhouses at the foot of the mountain.