Research On Citrus Recognition Method Based On Machine Vision

As one of the important origin of citrus,China has a history of citrus cultivation for more than 4000 years.Nowadays,citrus has become the fruit with the largest planting area,the highest yield and the largest consumption in China.In the citrus industry,fruit harvesting is an important link with high labor intensity and strong seasonality in the whole production process.The traditional manual harvesting method is not only inefficient,but also needs to employ a large number of labors,resulting in a large amount of cost.Research on the recognition method of citrus fruit based on machine vision can provide guidance for the automatic harvesting of citrus,which is beneficial to improve the efficiency of citrus harvesting and reduce the labor cost.The main research contents are as follows:(1)Design of Citrus harvesting manipulator based on machine vision.Based on the analysis of the working requirements and working environment of the harvesting manipulator,the overall structure design of the citrus harvesting manipulator based on machine vision is completed.The camera and other important hardware are selected and analyzed,and the control system is designed according to the work flow of the harvesting manipulator.(2)Research on Citrus image enhancement algorithm.To solve the problem that citrus targets in the images collected in the natural environment is difficult to recognize due to the influence of light,two effective citrus image enhancement methods are proposed.For the citrus image with uneven illumination distribution,based on the fuzzy set theory and combined with the global information of the image,the image pixels are divided into two categories: enhancement and suppression.Then the local information around the pixels is introduced to establish an exponential function to enhance the image.For the citrus image with low brightness,based on the Retinex principle,the v-component image after guided filtering is substituted as the illumination component,and the reflection component is separated by calculation.The improved bilateral gamma adaptive enhancement method is used to improve the brightness of the illumination component image,and the detail of the reflection component is enhanced by using the visual characteristics of human eyes.Both image preprocessing algorithms can effectively improve the brightness information of the corresponding citrus image.(3)Research on green citrus target recognition algorithm.To solve the problem that it is difficult to accurately identify green citrus targets in natural environment,this paper has proposed a saliency detection method based on manifold ranking algorithm.Aiming at the problem that the saliency graph of the traditional graph-based manifold ranking saliency detection algorithm is not ideal,a method combining relative total variation and local complexity is used to extract more accurate foreground seeds Finally,the extracted foreground seeds were combined with the boundary background prior saliency map to get the final saliency map.The improved method can effectively identify green citrus regions,and the accuracy is improved significantly compared with the former algorithm.(4)Research on Citrus target recognition algorithm in complex scene.To solve the problem that it is difficult to accurately identify citrus targets in complex natural environment,the regression based YOLOv3 algorithm was used to identify citrus targets.Establish the citrus data set with collected citrus images,and improve the learning and recognition ability of the network through pre training method.The YOLOv3 model is trained by using the training set,and the experiment and result analysis are carried out on the test set.The experimental results showed that the trained YOLOv3 model has good recognition performance in the test set data,and could effectively identify citrus targets in natural environment.