Research On The Detection Technology Of Tea Bud Based On Image

China is the hometown of tea.At present,tea planting,production,consumption and export are ranked first in the world.The wide demand for tea requires the strong support and development of tea picking industry.Nowadays,tea picking technology is mainly manual picking and mechanical picking.However,it is difficult to guarantee the efficiency and cost of manual picking and ordinary mechanical picking,which will also affect the quality of tea.Therefore,a technology that can automatically and selectively pick tea is needed,and the premise of this technology is the detection of tea buds.Based on the above background analysis,this thesis takes tea images as the research object to study and explore the method of automatic detection of tea buds.The main contents are as follows:First,the method of extracting color feature is used to detect tea buds.The collected tea images are processed by filtering and noise reduction,then the color features of tea images under the four color models of RGB,HSI,Lab and YIQ are compared and analyzed.After comparison,the G-B color features are extracted and grayed.For grayscale images,Otsu segmentation method is adopted to realize image segmentation.After morphological processing,the segmented binary images are masked with the original images,and the images with tea buds and a small amount of irrelevant information are obtained.Secondly,in view of the shortcomings of the methods for extracting color features,a new method for detecting tea buds is proposed,namely object detection algorithms based on deep learning: SSD algorithm and YOLO V3 algorithm.The SSD algorithm uses the improved VGG16 convolutional neural network to extract the features of tea images,and generates the default box of corresponding scale and aspect ratio according to the actual situation of tea images.Then it selects the positive and negative samples according to the NMS algorithm,and finally adopts the loss function to ensure the training.The YOLO V3 algorithm uses Darknet-53 to extract the features of tea images,and generates a candidate box similar to the default box in SSD algorithm.After that,it also uses NMS algorithm and loss function to control the training.The two algorithms are different in principles and characteristics,but both can be applied to the detection of tea buds theoretically.Finally,the two object detection algorithms are used to detect tea buds.The whole experimental process includes: establishing the platform Matlab visualization and deep learning platform,using the method of the crawler and camera to acquire tea images,using the method of extracting color features to realize the preprocessing of tea images,using four methods to enhance the tea image data sets,labeling the tea image data sets manually,and SSD and YOLO V3 algorithms training and adjustment parameters.In the end,the neural network model for detecting tea buds can be obtained.The results show that both SSD and YOLO V3 object detection algorithms can realize accurate and adaptive detection of tea buds after combining the preprocessing of extracting color features.In comparison with the two algorithms,YOLO V3 has higher accuracy and faster speed,which is more suitable for the detection of tea buds,and has greater support and help for the future automatically tea picking technology.