Fruit Recognition And Orientation Of Bell Pepper Based On Binocular Vision

With the continuous development and in-depth research of computer technology and deep learning,convolutional neural network has become one of the most cutting-edge and valuable research in the field of artificial intelligence,and the recognition of fruits and vegetables is the research hotspot of convolutional neural network in the field of agriculture.In order to realize the fast and accurate recognition and location of Lantern pepper fruit in natural scene,this paper proposes to use deep learning model combined with binocular vision to recognize and locate lantern pepper fruit.In the aspect of Lantern pepper fruit recognition,considering the complex factors such as light change,branch and leaf occlusion and fruit dense overlap in the natural scene,this paper selects yolov3 deep convolution neural network combined with efficientnet to recognize lantern pepper fruit.The detection network adopts efficientnet network model as the backbone network,which not only reduces the architecture and computational complexity,but also improves the efficiency and robustness of the network model.At the same time,the idea of multi-scale is introduced into the network to enhance the detection performance under different sizes.Through the training of yolov3 efficientnet model,the confidence of 0.5 is selected as the threshold for the detection of Lantern pepper fruit.The detection result accuracy P is 95.08%,the recall rate R is 85.70%,the average detection accuracy map is 95.79%,F1 score is 90.15%,and the model size is 61 mb.Compared with traditional recognition methods and other target detection algorithm models,the model proposed in this paper has higher detection accuracy Faster detection speed and good robustness.In terms of binocular vision,firstly calibrate the binocular camera to obtain the internal and external parameters of the binocular camera,then correct the binocular camera to make its optical axes parallel to each other,then match the features of the images of the two cameras by the stereo matching algorithm to obtain the parallax images of the two camera images,and then reconstruct the three-dimensional coordinates of the lantern pepper fruit by the parallax information and the calibration structure.Based on the principle of binocular vision and triangle similarity,the error range between the distance calculated by binocular vision and the actual distance measured is 0 ~ 15 mm,which basically meets the grasping accuracy.