基于RGBD的菠萝催花机器人花芯识别与定位

Flower core recognition and location of pineapple flower inducing robot based on RGBD

中文关键词  机器人  神经网络  识别定位  智慧农业

英文关键词  robotics  neural network  recognition and location  smart agriculture

中文摘要

      目前的夜间菠萝催花作业极度依赖人工劳作, 成本高昂且效率低下, 人工催花的工作模式制约了菠萝种植业的发展, 机器人和机器视觉的发展有助于高效完成菠萝催花任务. 针对夜间菠萝花芯识别存在夜晚光线不足、菠萝与野草混杂识别困难和定位不准确等问题, 本文提出了基于RGBD的目标检测算法, 用于菠萝催花机器人的花芯识别与定位. 本文算法首先通过基于HSV色彩空间的伽马函数自适应地校正夜间图像的不均匀光照; 其次, 使用YOLOv8目标检测算法实现菠萝花芯的快速识别, 得到花芯的预测坐标; 然后, 基于RGBD图像的深度信息实现了预测框的深度估计; 最后通过坐标转换得到花芯的空间坐标, 引导机器人完成催花作业. 本文算法在夜间菠萝图像数据集上进行验证, 花芯识别准确率为94.9%, 每秒可处理26帧图像, 满足实时催花作业要求

英文摘要

      At present, the night pineapple flower promotion operation is extremely dependent on labor, which is costly and inefficient. The working mode of artificial flower promotion restricts the development of pineapple planting industry,and the development of robots and machine vision help to efficiently complete the task of pineapple flower promotion.In order to solve the problems of pineapple flower core recognition at night, such as lack of light at night, difficulty in identifying pineapple and weeds mixed with each other and inaccurate positioning, an object detection algorithm based on RGBD is proposed for flower core recognition and location of pineapple flower inducing robot in this paper. Firstly,the proposed algorithm adaptently corrects the uneven illumination of the night image through the gamma function based on the HSV color space. Secondly, the YOLOv8 object detection algorithm is used to realize the fast recognition of the pineapple flower core, and the predicted coordinates of the flower core are obtained. Then the depth estimation of the prediction box is realized based on the depth information of the RGBD image. Finally, the spatial coordinates of the flower core are obtained through coordinate conversion, and the robot is guided to complete the flower prompt operation. In this paper, the algorithm was verified on the night pineapple image dataset, the accuracy of flower core recognition is 94.9%,and 26 frames per second can be processed, which meet the requirements of real-time flower prompt operation.