Research On Mushroom Detection And Monocular Camera Positioning Technology Based On YOLOv2 Network

The application rate of target detection,tracking,positioning and prediction in actual life scenarios is getting higher and higher,making its research value and actual use value rising linearly,and it has gradually become an important research direction of computer vision.The target detection,tracking,positioning and prediction systems can effectively reduce human labor and greatly improve work efficiency in industrial production.However,in actual application scenarios,the detection speed,accuracy,and real-time tracking of the target are still not high enough and the problem of target loss,which affects the subsequent positioning and prediction work.These problems will affect the performance of our target detection,tracking,positioning and prediction system.In order to solve these problems,this paper designs and implements a set of mushroom detection,Tracking,positioning and forecasting system for the actual scenario of sorting the mushrooms on the mechanized conveyor belt.First of all,this paper improves the currently more commonly used detection algorithm YOLOv2(You Only Look Once)algorithm.According to the characteristics of the slender mushroom roots,the shape of the detection frame is adjusted,and the horizontal detection density is increased.Then use ResNet50 as the YOLOv2 feature extraction network of the algorithm makes the detection accuracy better and can solve the gradient explosion problem caused by the network depth being too deep.Through experimental verification,after training with the same training set,the improved YOLOv2 algorithm has an accuracy of about ten percentage points higher than the unimproved YOLOv2 algorithm in detecting root-up mushrooms.It shows that the improved target detection algorithm used in this paper has certain advantages in detection accuracy.Then,the detection algorithm is used to predict the position of the detected mushroom on the conveyor belt at the next moment when the first frame of the mushroom with the root upward is detected to achieve the purpose of real-time detection.In order to obtain the real-time coordinate information of the mushrooms,the pixel coordinate information of the mushrooms with roots at this time is acquired at the time of detection and prediction,and then the pixel coordinates of the mushrooms with roots are converted into world coordinates according to the imaging principle of the camera.The real-time position coordinates of mushrooms are obtained.According to the real-time position coordinate information obtained,which is the sequence coordinates of the mushrooms,the short-term memory network(LSTM)is used to predict the upcoming position of the mushrooms on the conveyor belt outside the camera.After obtaining the coordinate information,you can control the robotic arm to grab the mushrooms.Finally,based on the above algorithm,this paper designs and implements a system for detecting,tracking,locating and predicting the mushrooms with the roots on the conveyor belt.The system can detect and track the mushrooms with the roots under the camera in real time.Positioning and forecasting.The system consists of a detection module for mushrooms with roots up,a continuous tracking module for mushrooms with roots up,a positioning module for mushrooms and a position prediction module for mushrooms.The system experiment proves that this system can effectively carry out fast and accurate detection and long-term tracking of the mushrooms with the roots on the conveyor belt.While detecting and tracking the mushroom with the root up,the position coordinates of the mushroom can be obtained and the position of the mushroom with the root up outside the camera can be predicted,which proves the feasibility of the system.