Research On Classification Of Small And Medium Crops In Ningxia Plain Based On Multi-source Multi-temporal Remote Sensing Data

Ningxia plain is one of the rice producing areas in China.The classification of crops in the rich areas of crop producing areas plays an important guiding role in crop yield estimation,food planning analysis and irrigation management.With the continuous launch of earth observation satellites,more and more remote sensing data are applied in the field of crop classification.Based on multi-source and multi-temporal remote sensing data,this paper classifies small and medium-sized crops in Ningxia plain.The main research contents and achievements of this paper are as follows:(1)An improved registration algorithm for optical and SAR remote sensing images was proposed.Image registration is an important step in image classification.In order to overcome the difficulty of heterogenicity image registration,in this paper,the local feature matching of optical and S AR images is carried out by combining the maximum extreme region detection method with the anisotropic filter to effectively remove speckle and noise of SAR images.The accuracy of the improved registration algorithm is nearly 20%higher than that of common algorithms.(2)The multi feature and multi time remote sensing image sequence is extracted and constructed.In view of the obvious limitation of single time image in the classification of land features with change information,this paper first obtains the multi images of a growth cycle of crops in the study area and the growth situation of different crops in each period-Secondly,the normalized vegetation index(NDVI),backscatter intensity coefficient,texture mean and other features of multi scene images are extracted,and the time series set is constructed to generate cart decision tree in the way of binary tree classification,which achieves better classification accuracy for different crops.(3)To analyze the effects of different classification methods on the classification of small and medium-sized crops.The unsupervised learning method k-means clustering,supervised learning method maximum likelihood method and random forest classification algorithm were used to classify the main crops of the research area,such as corn,rice and wheat.The experimental results show that supervised learning has better classification effect than unsupervised learning based on prior knowledge.Among them,the random forest algorithm achieves the best classification effect under the condition of multi-feature fusion,and the overall accuracy reaches 89.97%.The effectiveness of the method in the classification of small and medium-sized crops was verified.Based on a single high resolution remote sensing image classification for small and medium-sized pieces of crops has the problem of insufficient recognition rate is low,timeliness,through the extraction of optical and radar much time series of remote sensing image,the characteristics of the use of the growing of crops,and phenological phase of supervised learning and unsupervised learning classification methods,proves the multi-source long phase combination of remote sensing data in the effectiveness of the small and medium-sized piece of crop classification problem.