| Rice is an important food crop in our country,and its yield is of great significance to national food security.Rice is easily infected by diseases during its growth,which directly affects the yield and quality of rice.Rice diseases mostly occur in leaves,and the accurate identification of rice diseases is helpful for agricultural producers to implement precise control measures.At present,the identification of rice diseases mainly relies on the field identification by experienced agricultural technicians.This method requires high professional knowledge of agricultural technicians.It not only consumes a lot of manpower and material resources,but also has strong subjectivity and large error in the determination results.Therefore,it is of great significance to carry out research on the identification method of rice typical diseases and realize accurate diagnosis of rice diseases.Taking Jiuhong Modern Agriculture Demonstration Park in Qing ’an County,Heilongjiang Province as the experimental data acquisition base,image data of four different rice diseases including rice blast,striated leaf blight,red blight and bacterial brown spot were obtained.In this paper,the traditional machine learning method is used to realize the recognition of disease images,and the shortcomings of this method are analyzed and explained.Secondly,the classical convolutional neural network(CNN)framework is introduced to classify diseases,which effectively makes up for the shortcomings of traditional methods.Then,a lightweight CNN structure is proposed,which effectively avoids the shortcomings of large number of parameters and poor model performance in the calculation process of classical CNN.Finally,in order to realize the full information exchange in the CNN computing process and without reducing the classification performance and increasing the amount of CNN computing,the CNN structure with Interleaved Attention mechanism is further proposed.Finally,on the premise of considering the classification performance and the amount of convolution calculation,the accurate identification of four kinds of disease images was realized.The main contents and conclusions of the study are as follows: Firstly,the traditional machine learning method was used to extract the color features and texture features of the original diseases,and the extreme learning machine(ELM)and(RF)algorithm were used to identify them.The results showed that the RF data set obtained the best classification performance after the principal component analysis(PCA)dimension reduction of the fusion features,which reached 87.06%.Secondly,in order to highly integrate the process of feature extraction and pattern recognition,CNN is used for image recognition.After data enhancement,87.50% classification accuracy is obtained based on the improved Le Net structure,and the highest classification result is 93.33% for the four network models based on transfer learning.Then,in order to effectively reduce the number of parameters in the convolution process,ultra-lightweight subspace module(ULSAM)was introduced in this paper to reduce the number of calculated parameters,and 93.58% of the classification results were obtained,which improved the classification performance of the model while reducing the amount of calculation.Finally,the shortcomings of CNN convolution calculation were analyzed,and the CNN calculation method with the cross-attention mechanism was proposed,which could realize the information interaction between the convolutional channels and carry out the key convolution on the key features,and 96.14% of the classification results were obtained.The research results show that the deep learning method successfully realizes the recognition of four different disease images,and the research process effectively solves the problems of large number of parameters in the CNN calculation process,lack of information interaction between channels,and lack of key feature calculation.Compared with the classification performance of various methods,the CNN calculation method with the Interleaved Attention mechanism achieved the highest classification accuracy.This study provides a new idea for the efficient training of neural network in the case of small samples,and provides a reference for the image recognition and diagnosis of rice and other crop diseases. |
