Research On Tea Bud Recognition Based On Cotivolutional Neural Network

At present,manual picking is still the main method for tea picking,which requires a lot of human resources and is time-consuming and labor s.In recent years is given priority to with the depth study of target detection for the target positioning and target recognition has made great breakthrough,the Convolutional Neural network(Convolutional Neural Networks,CNN)not only has the ability of convolution computation,and have the characteristics of deep learning model structure,and the fusion image characteristics and characterization of learning ability,thus is widely used in the informationization of all walks of life.At present,the main problem for intelligent tea picking is machine vision-oriented information perception,location and recognition of the buds to be picked.Considering the status quo of intelligent tea picking and the development advantages of target detection,as well as the huge economic benefits of tea,this paper takes the budding of tea tree as the research object,and USES deep convolutional neural network to realize relevant research on intelligent tea picking technology.The main research contents are as follows:(1)Preprocess the target image interference factors,and use OSTU(maximum inter-class variance method)algorithm and threshold segmentation algorithm to segment the tea tree buds and obtain the contour image of tea buds.Experimental results show that the multi-threshold OSTU algorithm has a good segmentation effect for tea tree shoots with a single target,but the multi-target shoots segmentation effect is not ideal in complex environment.Traditional segmentation methods for the extraction of tea tree buds with complex backgrounds are subject to a lot of interference,resulting in very unstable segmentation results,which is not ideal for bud recognition in complex environments.(2)YOLOV3(You Only Look Once)model based on deep convolutional neural network was applied to the identification of tea tree buds in natural environment,and the identification of tea tree buds was studied from multiple scales.This paper discussed and analyzed the results of the first-order model YOLOV3 and the first-order model SSD(Single Shot Multibox Detector)under the deep convolutional neural network.The results showed that YOLOV3 was 9.1%higher than the Mean Average Precision value of SSD detection algorithm for complex tea tree buds identification,and the recall rate was 5.3%higher than that of SSD detection algorithm.Therefore,first-order YOLOV3 has a better effect on tea tree buds identification in natural environment than SSD detection algorithm.(3)Will identify the accuracy of high YOLOV3 YOLOV3 after the neural network model and the optimized-tiny neural network model and YOLOV3-SPP neural network model to identify the tea tree buds for,the mean and the average accuracy of recognition accuracy and time performance were analyzed,results show that the convolution of the neural network based on depth YOLOV3-SPP for bud recognition effect is better,the mAP of value is as high as 91%,highest YOLOV3-SPP model for the identification precision of the buds,YOLOV3 fastest-tiny model identification.The research results show that the application of convolutional neural network in the identification of tea tree buds in natural scenes is feasible,and the Yolov3-SPP model is suitable for the identification of tea tree buds,providing a technical basis for intelligent tea picking.