Research On Machine Learning In Rice Patch Image Recognition

Nowadays artificial intelligence develops rapidly in the world,and has infiltrated into all aspects of human life.As a big agricultural country,rice is our most important food crop.How to use machine learning to promote the development of agriculture has become a research hotspot.Based on the background of rice disease recognition and image processing technology,two different machine learning methods: Support Vector Machine and Deep Learning are studied,and the optimal model of rice disease identification is established by optimizing and improving the two algorithms.The main contents of this paper are as follow:(1)Four kinds of rice disease image data sets were established,Vector median filter,basic vector direction filter and distance direction filter were used to preprocess the image.By calculating the mean square error and peak signal-to-noise ratio on filtering effect evaluation,Vector filtering method is selected to filter the image at last.(2)Establish GrabCut image segmentation model based on significance analysis.In order to overcome the huge workload caused by manual frame selection of target pixels in traditional GrabCut segmentation method,the saliency algorithm of image is used to realize automatic target region extraction.Using the AC algorithm to get the global saliency and local saliency of the image,combine the two saliency images to get the final one.The fused saliency graph is designed as constraints and added to the Grabcut region item to realize Grabcut automatic segmentation.(3)Using improved support vector machine algorithm to realize rice disease image recognition.Firstly,use the color feature and HOG feature of the image to extract the feature of the lesion image,then the extracted feature is used as the input of support vector machine to train the classification model.In order to optimize the parameters of support vector machine,improve the generalization ability and classification accuracy of the model,particle swarm optimization was used to optimizethe parameters of support vector machine because of its global search ability.The trained model was used in rice disease image recognition process,and then the classification results were evaluated.(4)A DDC depth migration learning algorithm based on pre-training is proposed.Aiming at the shortage of samples in rice disease image set,the concept of migration learning was introduced,and a DDC depth migration learning algorithm based on pre-training was proposed.Pre-trained models on large datasets are migrated to rice image sets for retraining.There is a big diversity between different training samples,to solve this problem,a DDC algorithm is proposed.The final training results are compared with those directly trained on the training set,which verifies the effectiveness of the proposed algorithm and proves that the proposed algorithm has stronger representation ability.At the same time,in order to further improve the network performance,the original SoftMax classifier is replaced by PSO-SVM classifier.The results show that the network performance has been further improved.