Research On Fruit Classification Technology Of Dual-path Deep Belief Neural Network Based On Multi-feature Fusion

Deep learning constitutes a novel and modern technology for image processing and data analysis,which has great application prospect in image processing and data analysis,and the development of intelligent agriculture has also introduced deep learning.China is a large agricultural country with a long history,in which the area of fruit cultivation and output of agricultural products are among the world's leading,but the treatment technology after fruit picking restricts the competitiveness of the fruit market,so the realization of automatic classification and recognition of fruit species has become a necessary prerequisite for the modernization of domestic agriculture.Facing the demand,the study of fruit classification recognition technology starts from shallow learning and training,but such research methods need to rely on a large number of manual extraction of feature information,and there will be inter-class similarity phenomena between the fruit images,resulting in artificial extraction of characteristic information difficult to achieve the desired recognition effect.Based on the powerful automatic learning and training ability of the deep belief network model,this paper can efficiently complete the automatic extraction of multifeatured information,improve the recognition rate of fruit image,and expand the idea for the study of intelligent agriculture.The main work of this paper is as follows:(1)In this paper,the bilinear interpolation algorithm data set is used to complete the unification on the standard scale,and the color target body is obtained by using the automatic threshold segmentation elimination method interference.A deep belief network feature training model that combines texture features and color features is designed to perform multi-task learning in feature processing,expand fruit image features as many levels as possible,and achieve the effect of improving the distinguishability of the extracted features.In the color space model,the HSV model that conforms to machine vision is selected.In the process of texture feature image processing,the local texture is extracted.Choose multilevel texture extraction.First,through the Multi-Block Local Binary Patterns algorithm,and then use the improved Center-Symmetric Local Binary Patters to complete the second extraction,and finally the texture feature fusion extraction is done with the Histograms of Oriented Gradients algorithm.The fused texture feature image contains more detailed feature information.(2)In view of the problem of improving automatic learning and training of deep learning,and improving the traditional deep belief network model,a model based on double channel deep belief network is proposed.The color feature images of the training set are input into the first deep belief network model for training and learning,and the fused texture feature images are input into the second deep belief network model for learning and training,and then the color images and texture samples of the fruit image are independently learned and updated the network parameters,the two networks with different feature extraction capabilities are implemented,and the neural network fusion algorithm is used for the feature fusion of the top-level output feature vectors of the two independent networks to obtain a double channel deep belief network.(3)After completing the theoretical research,different kinds of fruit images have achieved high-quality recognition effect.In order to test the superiority of the scheme,the recognition results are compared with different training methods such as shallow learning models and traditional deep belief networks.The average correct recognition rate of the proposed recognition method is 95.9%,which is superior to other methods.At the same time,comparing the recognition effects achieved by different feature information under the same method,the recognition accuracy based on color features and multi-level fusion texture features is the highest.In summary,the experimental results show that the multi-feature fusion algorithm enhances the generalization of fruit image classification,and the double channel deep learning model improves the accuracy and robustness of fruit image classification.It provides theoretical support for the development of smart agriculture and opens up ideas for further research in deep learning.