Research On Tomato Recognition And Ripeness Detection Based On Computer Vision

In recent years,with the expansion of tomato production scale,the yield and quality of tomatoes have greatly improved.The traditional manual picking method is difficult to meet production needs due to the serious aging of the social population and the increasing shortage of agricultural labor.Therefore,achieving mechanization and intelligence in the tomato industry has become an urgent task.The rise of computer vision has made it possible for the tomato industry to become intelligent.This article uses computer vision algorithms to study the problem of tomato recognition and maturity detection.The specific research content is as follows:Firstly,in response to the different environments involved in the tomato picking robot picking process,this article selects two time periods for image collection: day and night.During the day,three different conditions of forward light,backlight,and backlight are selected for image collection,while at night,effective collection is achieved by adding auxiliary light sources,and the entire dataset is expanded through data augmentation methods.Then,select an image preprocessing algorithm to perform dim light removal on the nighttime tomato image.Finally,establish your own tomato image dataset for subsequent image recognition and detection.Secondly,for tomato recognition and detection issues,a large number of tomato images are collected as samples,and the images are annotated one by one using Labelme software to obtain a json format dataset.A tomato recognition and detection method based on Mask RCNN was proposed,and the structure of the Mask RCNN model was analyzed in detail and improved.The Inception module is introduced into the backbone network,and a new Ro I extractor is proposed in the RPN stage to replace the traditional Ro I extractor,and a hollow convolution algorithm is used in the network model.Then,the tomato detection model was experimentally validated on a self-made tomato dataset,and both AP and AR values were improved compared to the original Mask RCNN algorithm,The detection results showed that the improved algorithm not only achieved tomato recognition,but also achieved instance segmentation.Finally,for the maturity problem of picking tomatoes,a tomato maturity detection method based on computer vision is designed.(1)obtaining a target area(namely tomato fruit)through YOLOv3 algorithm;(2)Two feature extraction networks,RGB Net and HSV Net,are pre-trained,and on this basis,the extracted RGB and HSV color features are fused,and the classification network is used to realize tomato color feature recognition;(3)jud that maturity status of the tomato according to the correspond tomato color.The test results show that,compared with other classical network models,the tomato maturity detection method designed in this paper with color as the main recognition feature can realize the color recognition of tomato fruit with high recognition accuracy,and can effectively distinguish the tomato maturity.The above image preprocessing,tomato recognition,and maturity detection methods provide an effective solution for achieving intelligent tomato picking,which is of great significance for promoting the intelligent development of the tomato industry.