基于深度学习的花椒采摘器视觉识别

Visual Recognition of Pepper Picker Based on Deep Learning

对于智能花椒采摘器中机器视觉部分在花椒枝干识别与采摘定位上的不足,本文通过将深度学习技术中的卷积神经网络模型与注意力机制这两种模型运用到智能花椒采摘器的机器视觉部分以提高采摘器的识别功效。结果显示,经过优化后的卷积神经网络算法训练使采摘器对花椒簇的整体识别准确率由52.3%提高至96.7%,同时通过注意力机制算法提升了机器视觉对花椒树主枝干识别的抗干扰能力,帮助采摘器更加准确的判断出采摘点的位置。通过以上两种模型验证了深度学习技术在提高花椒采摘器机器视觉的算法准确性与抗干扰能力的有效性。

There are deficiencies in machine vision of intelligent pepper picker in the pepper branch recognition and picking positioning.This paper applies convolution neural network model and attention mechanism of deep learning technology to the machine vision part of the intelligent pepper picker,so as to improve the recognition effect of the picker.The results show that after the optimized convolution neural network algorithm training,the overall recognition accuracy of pepper cluster is improved from 52.3%to 96.7%.At the same time,the attention mechanism algorithm improves the anti-interference ability of machine vision for pepper main branch recognition,which helps the picker to judge the picking point more accurately.The above two models verify the effectiveness of deep learning technology in improving the algorithm accuracy and anti-interference ability of pepper picker machine vision.