Study On Image Recognition Method Of Corn And Soybean And Rice

In the new information era,computer machine vision technology has been applied in many areas as digital picture processing technique develops rapidly,and it has huge development potential in the future.Particularly in facet of search engine,traditional key word-based search cannot meet people's increasing demand,however,picture can reflects enormous information directly.Therefore,picture-based search becomes more significant.Picture identification is one of crucial technologies of picture-based search engine,and especially in agricultural industry,computer machine vision and picture processing technology have been utilized to identify and categorize blade increasingly.The research object of the article are northeast's major crops including corn,soybean and rice,while data collecting area is based on test field of Heilongjiang Bayi Agriculture University.Image of corn,soybean and rice is analyzed and understood by utilizing computer machine vision and picture processing knowledge,and identification and category of them are achieved ultimately.Following research is done for the experiment.1.First of all,understand growth knowledge of corn,soybean and rice.Picture gathering date and hardware are selected,and then picture of corn,soybean and rice is gathered.After that,picture is pre-processed by linear transformation and median filtering,and thus image's noise is lessened.Based on characteristics of the three crops,picture is segmented by K-means clustering method,so as to acquire target picture.2.The research identify and categorize feature parameter extracted.Before identification,first extract segmented picture feature parameter,and then use HSV color model to obtain picture's four color feature.After above,use Hu moment to acquire picture's seven shape characteristics,and utilize gray level co-occurrence moment to gain picture's nine texture feature.After that,use LLE manifold algorithm to reduce dimension of picture data and extracting Three-dimensional feature.Finally,use BP neural network,Elman neural network and FCM clustering algorithm to identify the feature vector.3.Convolution neural network is able to identify two-dimension picture directly.IThe convolution neural network is used to recognize the original image as input.The research shall optimize batch training samples,iterations and training set.When batch training sample is set at 10,and iteration at 100,correct recognition ratio is 100%.According to test,in the three methods,the recognition rate and training performance of BP neural network are better.Convolution neural network,among those algorithms,not only boasts high correct recognition ratio,but its picture can act as network's input directly.There's no need to pre-process picture and extract its feature.