Automatic Recognition And Segmentation Algorithm Research Of Fruit Image Based On Machine Learning

China is a big fruit-producing country,and its annual fruit output accounts for the world's top ranks.In fruit production operations,most fruit picking operations rely on manual operations.Due to the gradual shift of agricultural labor resources to other industries in the early20 th century,the shortage of agricultural labor force has become more and more serious.The fruit picking robot can solve the shortage of labor resources and increase the production efficiency.The automatic recognition and segmentation of fruit images is an important part of the visual system design of fruit picking robots.In order to achieve automatic fruit picking,this study proposes two fruit image automatic recognition algorithms based on SVM support vector machine algorithm and deep convolution neural network algorithm.After identifying the type of fruit,it is necessary to divide the fruit and the background to obtain the center of the fruit.This study also proposes an image segmentation algorithm based on BP neural network and morphology processing,and an image segmentation algorithm based on color difference and gradient feature.In the fruit image recognition,machine learning and deep learning techniques have a wide range of applications.In this study,the SVM support vector machine algorithm and deep convolutional neural network algorithm in machine learning are used for apple,orange,mango and other six different types of fruits.The image is automatically recognized and the two algorithms compare each other's advantages and disadvantages.The machine learning SVM algorithm does not require a large number of samples,and the training time is short.However,it takesa lot of time to artificially extract representative image features.The accuracy of the algorithm is finally 75.35% in its own data set.The deep learning algorithm migrated Google's Inception-v3 deep convolutional neural network model,and made relative changes to the network architecture for its own data.This algorithm requires a large number of training samples,but the biggest advantage is that there is no need to manually extract features.The convolutional neural network will find its own effective features,even some very important high-dimensional features that cannot be extracted by humans.After 40,000 iterations of the algorithm,the recognition accuracy of the model on the test set is93.9%.The recognition rate has greatly improved compared to the SVM algorithm.In the segmentation of fruit images,two different image segmentation algorithms are proposed in this study.The purpose is to segment the image of the fruit part from a complex background and output the center of the fruit.It is an image segmentation algorithm based on BP neural network and morphological processing,and an image segmentation algorithm based on color difference and gradient features.The BP neural network takes the H channel component value of the 3×3neighborhood of the HSI color model image as the image feature,and uses Photoshop software to segment the training sample as the expected output.The number of iterations is 1000 and the error is 0.001.Get a reasonable model.Aiming at the shortcomings of traditional threshold segmentation method for segmentation of fruit gray image,a fruit image segmentation algorithm based on color difference method and gradient feature was proposed.The algorithm uses the absolute value of the color difference of the R and G components in the image color feature and the edge gradient feature of the image,and uses the largest cluster-likevariance method to segment the image.Both algorithms achieve accurate segmentation of the original image,and the error in the center position is within 10 pixels,and the center position of the fruit can be accurately obtained.