Wheat Map Features Research Based On Deep Learning

Wheat maps are an image set of various wheat varieties,it plays an important role in the research of wheat related characteristics.The classification of wheat varieties has important guiding significance to the optimization of wheat industry and food processing.Due to the shortcomings of the main detection methods such as chemical analysis and empirical learning,it is urgent to introduce a novel and fast non-destructive method for wheat detection.In this paper,with the classification accuracy of wheat varieties which was used as the research target,some research work related to wheat features was carried out.The collection and treatment of wheat maps are an important link in the study of wheat features.Nine wheat varieties were collected in this paper,including DNS,NS,Shiluan 02-1,Zheng 366,Hungary,APH,Jinan 17,Zhouyuan 9369 and Yongliang 4.In this paper,the collected wheat images are labeled by hand,and the maps set is packaged into binary,which improves the rate of computer operation.In addition,this paper applies the methods of data balance and data expansion to the preprocessing of maps set.It is beneficial to enhance the expression ability of wheat maps.According to the principle of statistical features of wheat and the principle of deep learning,two kinds of study about wheat features are designed in this paper.Firstly,by extracting the statistical features of color,morphology and texture of wheat,a heuristic scheme of wheat feature analysis is designed in this paper.The relationship between the statistical features of wheat and the effect of wheat classification is studied by using the complete gradient clustering algorithm.The results show that the effects of statistical features on wheat classification are in order of morphology,texture and color.As the number of wheat varieties increased,the effect of statistical features on wheat classification decreased,so the method of wheat statistical features was not applicable.Secondly,according to the feature extraction ability of deep learning,a wheat feature analysis scheme based on convolution neural network is designed.In this paper,the optimal network structure of nine wheat classification can be determined by contrast experiments,and the generalization of the network model is improved by data preprocessing method.The experimental results show that the convolution neural network method can extract wheat features more deeply,and the classification effect of wheat is better.The model has better generalization performance by using pre-treated data training model,and the experimental results visualize the variation of wheat features in the network layer by layer.In this paper,the two experimental schemes have different results on the same wheat maps.The results show that the traditional research methods with wheat statistical features are not applicable and can not shoulder the complex and changeable task of wheat classification.In addition,the experimental results also show that the deep learning method has a advantage of feature extraction and classification performance of wheat,and has stronger applicability and generalization ability.