Crop Disease Identification Based On Video Monitoring

Intelligent and accurate identification of crop diseases is the basis for modern diseases prevention and control,and an important guarantee for improving crop quality and yield.Video monitoring has the advantages of good dynamics and high cost performance,which makes up for the corresponding deficiencies of static image recognition and spectral imaging methods.However,there are many problems such as huge redundancy of video information,poor extraction of disease areas and poor disease recognition performance,which lead to the stagnation of application and promotion.In this paper,tea plants are taken as the research object.Using machine vision and deep learning technology,including key frame extraction,saliency map extraction and CNN identification methods to monitor crop diseases intelligently from videos which contains abundant and comprehensive crop growth information.The chief content and result are listed as below.1)Research on key frame extraction of crop monitoring video.In order to extract key frame from the monitoring video accurately and efficiently,a key frame extracting algorithm which was based on optimal distance threshold clustering and feature fusion expression was proposed.Firstly,in order to obtain the frame class image set with optimal clustering number,we analyzed the difference between frames of the video,and determined the optimal distance threshold which is used for unsupervised clustering of inter-frame distances.Secondly,in order to extract the representative frame of each cluster,we calculated and merged color complexity and information entropy,and extracted representative frame based on 'cluster average' concept.Lastly,representative frame extracted from each cluster was assigned to the key frame image set.This algorithm solves the problem of the dependency of unsupervised clustering on the threshold,taken moving target changes and environment anomaly into account,and has good performance and adaptability.2)Research on the extraction of tea plant leaf diseases saliency map.In order to improve the extraction of tea plant leaf disease saliency map under complex backgrounds,a new algorithm combined SLIC(Simple Linear Iterative Cluster)with SVM(Support Vector Machine)was proposed.Firstly,super-pixel blocks were obtained by SLIC algorithm,significant point was detected by Harris algorithm,and fuzzy salient region contour was extracted by convex hull.Secondly,the four-dimensional texture features of super-pixel blocks in salient region and background area was extracted,then classify the super-pixel blocks by SVM classifier and the classification map was obtained.Lastly,the morphological and algebraic operations were used to repair classified super-pixel blocks,further the accurate saliency map of tea plant leaf disease images was obtained.The result shows that the proposed method can extract tea plant leaf disease saliency map from complex background effectively and has great significance in the study of tea plant leaf disease identification.3)Research on the identification of tea plant disease with small samples.Three kinds of common and similar tea plant diseases image including pestalotiopsis theae,tea anthracnose and tea brown blight was identified by the convolution neural network(CNN)under the condition of small samples.Seven preprocessing modes were designed and used to process original tea plant leaf disease images automatically.The classic AlexNet network model was used to carry out the learning tests,the training and actual effect was compared and analyzed.This research indicates,in the case of small samples,the proposed pretreatment method can effectively distinguish and identify three kinds of characteristic similar disease with high recognition accuracy and good performance.