Leaf Disease Recognition And Quality Prediction Management System Development For Apple

In recent years,the effect of climate change on apple quality is gradually increasing.Scientific quality prediction based on meteorological factors plays an important role in farmers' income,fruit supply and demand,market circulation and so on.Apple diseases often occur,but because fruit farmers lack disease identification and control techniques,it is easy to misjudge the types of apple diseases and misapply pesticide,resulting in economic losses.Early detection and diagnosis of disease types and precise prevention and control have an important role on guaranteeing apple quality in China.With the continuous development of machine learning and deep learning,it is possible to predict scientifically and effectively by means of algorithms.Aiming at the aspects of apple quality prediction,image recognition model and so on,the related work has been carried out.The main research contents and research results are as follows:(1)To solve the problem of long training time of stacked sparse autoencoder,K-means clustering optimizing deep stacked sparse autoencoder is proposed.The model has two improvements: traditional deep stacked sparse autoencoder introduces K-means algorithm and parallel cascade learning mechanism.The former reduces the input data scale of parallel sparse autoencoder by clustering input vectors of input layer and hidden layer into K clusters and inputing them into K sparse autoencoder in parallel for training.The latter effectively integrates the weights of parallel sparse autoencoder through parallel cascade learning to accelerate the running speed of the model.Compared with the stacked denoising sparse autoencoder model proposed recently,this improved model has higher recognition accuracy,greatly reduces the pre-training time of the model and speeds up the development of the model.The experimental results show that the traditional training method has redundant training phenomena,and the small sample set partition could help the network learn better parameters.(2)Aiming at the problem of over-fitting of transfer learning mechanism and image enhancement method on small sample data sets,mixed image amplification method is proposed,that is,noise-free pictures and noise pictures is input into the network for training to alleviate network over-fitting.Experiments show that in small sample datasets of apple leaf disease,mixed image augmentation makes the model get higher recognition accuracy.Secondly,a quality prediction model applying support vector machine based on simulatedannealing and particle swarm optimization algorithm is established.The model can effectively avoid premature phenomenon by introducing simulated annealing to optimise the positions of the particles locally in each iteration process,but its convergence performance is general and more iterations are needed.The support vector machine model based on particle swarm optimization has fast convergence speed,but it is easy to fall into local minimum.Comparing comprehensively,it is better to use the support vector machine algorithm based on simulated annealing and particle swarm optimization algorithm to construct the apple quality prediction model.(3)The apple leaf disease identification and quality prediction management system was developed.In order to realize the timely diagnosis of apple diseases and the scientific prediction of apple quality,the system designs data acquisition layer,data storage layer,data mining analysis layer and data information and result visualization layer,completes the construction of the overall system architecture and functional modules of the apple management system,excavates the collected information,and realizes the apple basic data subsystem,apple quality prediction subsystem and disease image recognition subsystem,which can provide better auxiliary decision services for fruit growers and the government and provide the support for the precise prevention and control of apple diseases.