Application Of Deep Learning Network In Recognition Of Apple Surface Lesions

Deep learning is a new field in the machine learning research.It mimics human brain mechanisms to interpret data,such as images,sounds and text.With the mature development of deep learning,deep learning has been widely used in various fields.In order to classify the mature apples,this paper studies the application of deep learning to the recognition of apple lesions.First of all,collect samples of apple images.It can be divided into four types,intact,rot,bug and crack and then we use the images to establish image database of apple surface.Then,the size reduction,gray level and the reduction of the dimension of the image are processed according to the different deep learning methods for the data input format requirements.After building the apple image database,this paper proposes the image recognition method of apple lesions based on convolutional network which uses traditional convolutional neural network to recognize and classify the apple images.At the same time,the traditional convolutional neural network is improved and optimized to shorten the training time and improve the classification accuracy.Secondly,a hybrid model based on the combination of depth belief network and support vector machine is built to classify and recognize apple images.Finally,we use convolution restricted Boltzmann machine to build a deep belief network to recognize apple images.The main work of this paper includes:(1)This paper describes the development process of deep learning,and analyzes the advantages of deep learning network compared to shallow network.This study uses traditional convolutional neural network to classify the collected apple images,and optimize traditional neural network.We combine the improved neural network with support vector machine to complete the task of image recognition and image classification.The experiment results show that the image feature extraction and classification results are both improved.(2)Several restrictive Boltzmann machines and the support vector machine are connected to construct the multi-layer classification model,and the apple image is identified and achieved good results.Among them,feature extraction uses deep learning algorithm,and feature classification uses support vector machine.Make full use of this classification model to identify apple image.It is found that the classification model can effectively recognize the images in the process of experiment.Through continuous trainings,we find how the number of samples,the number of network layers and the number of nodes effect on the accuracy of image recognition,and get the relationship between the various parameters and the accuracy of image recognition.(3)Deep learning network is constructed of convolution restricted Boltzmann machine,and in the training process of the network,we use the unsupervised learning and supervised learning alternately.Using the improved deep learning network which built by convolution restricted Boltzmann machine to recognize apple images.It is proved that the network is efficient and available,and the relationship between the convolution kernel and the recognition result is verified.