Kiwifruit Recognition In Natural Scene

China is one of the largest countries in kiwifruit yields all over the world,however,the handpicking work is a time-consuming and laborious,with the development of science and technology,automatic picking technology has gradually seep into people’s production and life.Accurately identifying the kiwifruit in natural scene is becoming a primary task of picking robot,so the research of kiwifruit recognition methods is necessary.In this paper,the image of the kiwifruit was segmented firstly,and secondly features extraction was performed on the feature of artificial designing and deep learning.Finally,the recognition model was established by using neural network and support vector machine classifier,and the extracted feature vectors were used as input parameters which provided a new method for the kiwifruit recognition.The main research contents of kiwifruit recognition based on computer image analysis are as follows:(1)To analyze the image of kiwifruit samples obtained,the method selected the Otsu threshold method based on the chromatic aberration method and Renyi entropy threshold segmentation method to segment the fruit and the background.After comparison and analysis of experimental results,this paper finally choose R-0.9G,G-0.9R,G-R,R-G color channel to segment fruit,leaf,grassplot,sky and trunk.As for part of the background which is complex and contains numerous fruit,Renyi entropy threshold segmentation method was used.For binary image to remove different types of residual noise also used the method of morphology to processing which had a solid foundation for features extraction.(2)The target object’s contour was drew and the target image was extracted by the minimum external rectangle algorithm.According to color features and special kiwifruit epidermis burr consisting of texture feature,this experiment to extract HSV color features,Gray-level cooccurrence matrix characteristics and Tamura features in contrast,linearity and roughness.Those 10 features and deep learning characteristics based on PCANet which had a solid foundation for kiwifruit recognition.(3)According to the characteristics of sample,the identification model was established based on neural network and support vector machine.We did the experiments about BP neural network,PCANet deep learning model and support vector machine algorithm.Experimental results showed that the recognition rate of PCANet was 94.92%,SVM was 87.67% and BP neural network was 65.09%.In conclusion,through the analysis of the characteristics of kiwifruit sample,a method of using artificial design features and deep learning characteristics to represent the kiwifruit sample was proposed and using neural network and support vector machine classifier to identify the kiwifruit and background had excellent results.The recognition of kiwifruit is realized accurately.