The Algorithm Research Of Fruit Image Recognition Based On Deep Learning

Since image recognition is a typical application in the field of pattern recognition, how to identify the image quickly and accurately has been an important topic for scholars in many countries. Fruit image recognition plays an important role in the field of intelligent agricultural and digital medical. In the wisdom agriculture, the recognition of fruit can achieve precise cultivating compound orchards of fruit trees and fruit of automated picking;as for digital medical, identifying the fruits mainly used for analyzing of the nutrition composition of auxiliary late fruit, which can help patients to establish a reasonable diet.However, how to fast and exact recognition of fruit is the keynote to these works. The current general fruit image recognition method cannot meet the needs of application, so we need to find a more effective algorithm for identification of fruit images.As a discipline in emerging and booming in the field of machine learning, deep learning not only changes the traditional machine learning methods, but also affects our understanding of human perception, which has been widely used in the field of image recognition and speed recognition. Therefore, based on the in-depth study of deep learning theory, deep learning is applied to the fruit image recognition to improve the fruit image recognition performance. This paper including:1. This paper briefly introduces the traditional method of image recognition and its existing problems, summarizes the research status and development tendency of deep learning, comparative analyzes the advantages of deep learning relative to the shallow learning and introduces two commonly used methods of deep learning training process in detail.2. Fruit image recognition algorithm based on Convolutional neural networks has conducted in-depth research. Taking into account that the different activation functions and sampling methods have great effects on network performance, we select appropriate activation function and the sampling method through the experiment, and discuss theactivation function and the sampling method. In order to further improve the network performance, we expand the network depth of fruit image recognition. Experimental results verified dataset in the network performance has a certain increase as the depth increases.Finally, we compared the fruit image recognition based on Convolution neural networks with traditional fruit image to show the effectiveness of the methods.3. Aiming at the limitation of the convolution neural network long training time, this paper designed the fruit image recognition method based on the deep belief networks. In order to solve the disadvantages that the deep belief network ignores the image of local structure and is difficult to learn about image local characteristic of faults and consider the fruit image affected by illumination change, we choose a deep belief networks based on the Census transform for fruit recognition. First, extract the texture characteristics of fruit through the Census transform to eliminate the influence of illumination changes on the feature extraction. Census transform can effectively retain local characteristics, which makes deep belief networks can effectively learn to image local characteristic, while reducing the unfavorable characteristics of deep belief network. Second, by using fruit image Census characteristic training depth belief network, get related network parameters.Finally, use deep belief networks for fruit image recognition. The experimental results show that this method exhibits the characteristics of strong learning ability. The recognition performance is superior to the conventional recognition algorithm and achieves the convolution neural network recognition effect and the training time was greatly shortened.