Research On Watermelon Flower Body Recognition Based On Machine Vision

Machine vision technology has been widely used in the field of agricultural engineering,mainly focuses on the extraction and recognition of crop morphology,color,size and shape,in order to detect crop growth,disease and insect pests.At present,there are serious problems in greenhouse crop pollination,such as large labor demand,bad environment and the decline of yield or quality caused by artificial pollination.If the target crop can be identified by machine vision,and then provide technical support for its intelligent pollination equipment,it will have broad application prospects.In this paper,watermelon flower body was taken as the research object,and machine vision recognition was carried out for watermelon flower body characteristics,posture and pollination status.Firstly,the watermelon flower images collected from natural environment were classified and analyzed,and the image database of watermelon flower posture and opening period was established.Secondly,a watermelon flower segmentation method based on Lab color space is proposed,prepossess watermelon flower image(image filtering,edge detection,histogram equalization)and converts it to Lab color space to analyze three-channel component image.the b channel complete image segmentation by OTSU algorithm.Then the machine learning method is studied.The feature extraction of watermelon flower sample image is carried out by using image feature recognition method,and the classification and recognition are carried out by SVM.Compared with the automatic feature extraction and classification based on Faster-RCNN neural network.The results show that the recognition rate of single state based on image features is better,especially the recognition rate of forward pollination state and flower character reaches 70%,and the recognition rate of multi-state fusion based on Faster-RCNN neural network can reach about80%.Finally,an embedded watermelon flower recognition and processing system isdesigned.The system relies on raspberry PI 3B + and Horned Sungem as hardware architecture,compatible with Faster-RCNN neural network algorithm,and carries neural network acceleration chip to improve the overall recognition speed.In this paper,machine vision technology is used to identify the flower state of watermelon.The embedded recognition system designed has high accuracy and wide applicability.It provides technical support and research basis for intelligent pollination device.