Research On Counting And Disease Diagnosis Of Wheat Based On Machine Learning

Due to the impact of global environmental changes and population trends,crop yield and quality are facing severe challenges.The key to solve the problem is to promote the overall intelligent,precise and efficient pattern of agriculture.As an efficient data and image processing method,machine learning adapts to the requirements of agricultural big data environment and intelligence.It can accurately improve the intelligence level of each link of agriculture in the production environment,assist humans to better meet the realistic needs of agriculture and crops,and make more efficient use of and enhance the value of natural resources.Wheat is one of the widely cultivated crops,which is particularly important to ensure the stability of food security in China.In this paper,the phenotypic information of crop wheat is taken as the research object.Aiming at the problems of yield prediction and disease diagnosis in the field of wheat production,the application of machine learning in wheat ear counting,disease recognition and severity estimation is deeply studied.The main research results are as follows:(1)An image frequency domain decomposition method combining multi-scale support value filtering(MSVF)and improved sampled contourlet transform(ISCT)is proposed for wheat ear counting.Support Vector Machine(SVM)is the classic classification and regression algorithm of machine learning.MSVF based on this has strong frequency domain filtering and generalization ability,which can effectively remove the complex background,while the multidirection characteristics of ISCAT enable it to represent the contour and texture information of wheat ears.In order to improve the level of wheat yield prediction,MSVF-ISCT method is used to decompose the ear image in multi-scale and multi direction in frequency domain,reduce the interference of irrelevant information,and generate the sub-band image with more abundant information components of ear feature information.Then,the ear feature is extracted by morphological operation and maximum entropy threshold segmentation,and the skeleton thinning and corner detection algorithm are used to count the results.The number of wheat ears in the image can be accurately counted.Experiments show that compared with the traditional algorithm based on spatial domain,this method significantly improves the accuracy of wheat ear counting,which can provide guidance and application for the field of agricultural precision yield estimation.(2)A dual attention mechanism classification network model(GR＿CSCNN)based on global response was proposed for wheat disease recognition.The model according to the distribution characteristics of disease characteristic information,the network graph generated characteristics,respectively from the dimension of channel and spatial build attention module,adaptive to establish correlation between global and local characteristics of the information,to generate the weights to enhance disease characteristics of information,and can be used as a simple and easy to use module integrated into tasks to end-to-end network architecture study.In addition,the network uses the global response idea for reference,and obtains the interdependence between pixels through matrix dot product operation,so as to avoid the complex operation such as manual design of pooling layer.Experimental results showed that compared with the traditional disease recognition methods,the network model could model the interdependence relationship on the local features of channels and space,which significantly improved the accuracy of wheat disease recognition.Moreover,the ablation experiments showed that the effect of the network model was better than that of no addition or single attention mechanism.(3)A convolution neural network model based on recurrent spatial transformer(RSTCNN)was proposed to estimate the severity of wheat leaf diseases.Aiming at the problem that the difference of color and texture features between images with different disease severity is small,and the accuracy of traditional disease recognition methods for disease severity estimation is not high,RSTCNN designs a multi-scale network model based on the improved basic classification sub network of spatial pyramid pooling as the framework.In the training process,the dimension of feature map is fixed,and the feature extraction and classification are realized at the same time.The region detection sub networks are connected,and the attention region in the feature map generated by the superior scale network is cut out through spatial transformation,which is used as the input of the lower scale network.Through the joint optimization of intra scale classification loss and inter scale ranking cross entropy loss,the recursive learning of attention region detection and fine-grained feature expression is realized in the way of alternating promotion.After normalizing and unifying the output of each scale network,the final result is obtained by integrating the features into the classifier.Experiments show that the network model can effectively estimate the severity of wheat leaf diseases.In summary,the research of wheat phenotypic information recognition based on machine learning solves the problems of wheat ear count,disease recognition and severity estimation in wheat yield prediction and disease diagnosis of agricultural production field,and adapts to the scientific and technological requirements of smart agriculture,which has high reference value and broad application prospects.