基于激光成像的棉花中白色异性纤维检测新方法

Foreign fibers,such as polypropylenes, plastic film are almost invariably present in cotton during the cotton collection and transportation. The existing machine vision systems are capable of detecting most of them by their color differences in visible lights，except non-fluorescent white foreign fibers. In the project,based on cooperative imaging of laser, we will study the methods for detecting the typecal white foreign fibers from the cotton in different processing stages, especially the transmissive and reflective properties of their surface under the illumination of laser line-structure lights, as well as the cooperative imaging methods in multi-angle of view for the sample multi-postures when they fly in a transparent tube with different sections in the size,shape and direction. Then, based on the cooperative imaging system which built by multi-programmable image sensors, we will design the algorithms for selection and identification of the suspecious target images from the sequence images and removal of the redundant images, as well as the algorithms for detecting the non-fluorescence white foreign fibers in machine vision inspections. The project is significant for cotton production and tixtile industry in our country, as the outcome of the research would reduce the system cost and significantly improve the detecting rate of the foreign fibers.

棉花在收集、运输的过程中容易混入化纤、塑料薄膜等异性纤维（foreign fiber）。现有的机器视觉系统在可见光照明下根据颜色区分棉花和大部分异纤，不能识别非荧光的白色异性纤维。本项目根据申请者的初步激光实验，研究基于激光成像的棉花中白色异性纤维的检测技术，特别是在不同波长、不同功率的激光照射下不同工序、品种的棉花和典型白色异纤的透射、反射特性和形态特征;以及样本在变尺寸、变轨迹、变速度的空气通道中的多姿态运动及其多点、多视角协同成像的原理。然后用多个可编程图像传感器构建协同式成像实验系统，研究从序列图像中筛选目标图像、去除冗余图像的可疑图像的判别方法；以及基于协同成像的非荧光白色异性纤维的机器视觉检测算法。本项目将在降低成本的前提下，大幅提高白色异性纤维的识别率，对我国棉花生产和棉纺织业的发展意义重大。

针对棉花中白色异性纤维检测的难题，本课题提出了一种基于线激光成像的机器视觉检测方法。首先，以籽棉、皮棉和棉花中常见的白色异性纤维为实验样本，研究得出了棉花和白色异性纤维的可分度与激光波长、功率、曝光时间之间的关系曲线，提出了激光成像的最佳参数。在此基础上，研究了线激光照明引起的棉花和异性纤维温升规律，为检测方法的安全性提供了保证。其次，鉴于深色异性纤维吸光性强和反光性弱，本项目研究了基于多个LED和线激光照明、多个低成本面阵相机的分布式成像检测系统，提出了LED和线激光双光源混合照明、采用线扫描相机一次成像检测的方法。基于上述研究得出的激光成像方法，本项目提出了基于面阵相机的异性纤维检测算法和基于线扫描相机成像的异性纤维检测算法。最后，基于深度学习、人工智能的最新进展，本课题提出了籽棉中异性纤维检测的最佳成像方法和基于CNN的人工智能识别方法。本项目的研究已经发表学术论文6篇，其中SCI期刊论文1篇，EI期刊论文5篇，发明专利实审2项，实用新型专利2项。已毕业博士研究生1人，硕士研究生7人。