| With the gradual enrichment of daily diet,the types of foods that use tomatoes as raw materials are also increasing.In the era of increasing shortage of labor force,artificial picking tomato takes time and labor,and the production cost is very high.We should develop intelligent picking equipment to improve picking efficiency.Order to solve the problems of inaccurate identification of tomato aims and inability to locate the picking point of tomato string,this paper carried out research on target identification of tomato aims in greenhouse and the location of the picking point of ripe tomato string based on deep learning.The main research contents are as follows:(1)The tomato aims target detection technology based on improved Cascade RCNN network was studied.In order to improve the recognition ability of overlapping fruits,the non-maximum suppression algorithm in the original Cascade RCNN network is improved to a Soft-NMS algorithm that is conducive to identifying overlapping targets.And then,the improved network is compared with other networks,and the results show that the improved network can accurately identify overlapping tomato fruits.compares the improved network with other networks.(2)A semantic segmentation technology of mature tomato string images based on improved ENet network was studied.In order to increase the network’s ability to extract detailed information,this paper adds a dense cavity convolution module between the encoder and decoder of the ENet network.The comparative experiments of the improved ENet network,U-Net network and Deeplab-V3 network show that the improved network can effectively segment ripe tomato bunches.The preconditions are prepared to determine the picking point of tomato skewers.(3)The position of tomato string picking point was researched by binocular camera,and got the three-dimensional coordinate of picking point.The two-dimensional coordinates of plucking point of tomato string were got by using Shi-Tomasi arithmetic,and then Taking pictures of tomatoes with the calibrated camera.Finally,the three-dimensional coordinates of picking point were calculated according to the principle of binocular camera.(4)Based on the QT development library,the tomato identification and tomato string picking point positioning software is made,which can realize the function of tomato fruit identification and tomato string picking point positioning in this paper. |
