An Automatic Identification System Of Ramie And Cotton Fibers In Longitudinal View Based On Image Processing

At present, manual recognition method is widely used in commodity inspection system for determining the blending radios of ramie and cotton blending textiles. This method is mainly dependent on the subjective judgment of operators; therefore, the experience and technique of operators will influence the objectivity and accuracy of result. Furthermore, the manual method is time-consuming, hard sledding and has a lower efficiency. Accordingly, commodity inspection system needs a kind of objective and accurate recognition method for identifying ramie and cotton. It is also the purpose of our research to realize an automatic identification system of ramie and cotton fibers in longitudinal view.This paper introduces an automatic identification system, which can identify ramie and cotton fibers and analyze compositions using textile fiber longitudinal image, based on the image processing and pattern recognition technology. In our research, a new method for identification textile fiber type using characteristic difference analysis of the stripes on fiber surface was established to substitute the manual method used widely. An automatic system without manual intervention was realized, which can identify fiber type objectively and accurately, measure fiber width and count fiber, as well as convert the data into the blending radio of ramie and cotton using the national standards used in commodity inspection system. The experiments showed that the overall tolerance for false identification of cotton or ramie fiber was lower than 7%.The primary contents of the research include: System hardware building and automatic control realization, including microscope system renovation, the fundamental and complicated controls of 3D microscope objective table; Utilizing the "C++" language to compile system software, and utilizing multi-thread technology to realize the collaborative work of the 3D objective table movement control part, digital camera image capturing part, and image processing analysis and fiber type recognition part; Designing perfect flow for ramie and cotton fiber recognition, then obtaining available characteristic parameters for the identification and analyzing probability distribution curves of the characteristic parameters, at last establishing an equations for the identification and determining the weight coefficients of the equations; Testing and analyzing the automatic system.The hardware of system includes microscope system, digital camera and 3D automatic objective table. The premise of realizing the automatic system is based on the completely automatic hardware platform which can be controlled by system software. Therefore, the ordinary microscope was transformed by replacing the original manual objective table with the 3D automatic objective table, the fine movement of which can be controlled by software through the R232 serial port; Digital camera was connected with USB2.0 to realize dynamic image capturing. Therefore, the system has the functions of 3D automatic objective table movement and image capturing simultaneously.Based on the hardware platform, this paper elaborated the system principle and realization method by two essential steps: fiber image locating capture and automatic identification of ramie and cotton fibers.The two essential steps have realized following functions: based on the established 3D automatic objective table micro hardware system to realize fiber image automatic capturing, panoramic scanning, automatic focusing, fiber image locating capture and other complex controls of the software and hardware; stitching and analyzing panorama to locate fiber targets on the sample slide and capture the fiber images without missing and duplicating any fiber segments; proposing characteristic parameters for identifying ramie and cotton fibers in longitudinal view, which can give perfect qualitative describes of the difference on fibers surface; obtaining probability distribution curves of these characteristic parameters, and establishing fiber type identification equations based on principle of the biggest probability; gaining weight coefficients of the equations through iterative auto-adapted identification tests. The final identification equations were utilized in the system and the overall tolerance for false identification of cotton or ramie fibers was controlled in an anticipated scope.The paper has also discussed some quite important questions for system realization thoroughly and given corresponding solutions as follows: correlation analysis of the six characteristic parameters used in the system; the principle of the identification equations; locating capture method for selecting perfect fiber image after comparison with another method; using breaking and reconnection to repair fiber skeletons, which enhanced the accuracy of the panoramic analysis and fiber image locating capture; solving the problems such as heterogeneous type processing for fiber skeleton overlapping, the deletion of skeletons branch, etc.After the automatic identification system was realized, we analyzed and tested validity, stability and reproducibility of the system. More than 30,000 ramie and cotton fibers from three provinces in China and from eight different countries, respectively, were identified in our system automatically, and the experimental results showed that the overall error for identifying cotton or ramie fiber was lower than 7% in this system, and the experimental results also confirmed that stripes analysis of fiber surfaces was an effective method for automatic identification of ramie and cotton in a longitudinal view.In addition, the test on the stability of the system indicated many different type blending fibers displayed the high stability. The same specimen has been automatic identified by the system at different time, and the result demonstrated the high reproducibility of the system.This paper also discussed the factors affecting the identification rate of the system: the examination speed, the optimum length of the fiber samples, the least number of fibers required for testing blending ratio, as well as several questions still existence in the system.Combining digital image processing and pattern recognition technology, the paper proposed a method to distinguish ramie and cotton fiber type utilizing fiber stripe analysis, and realized an automatic identification system of ramie and cotton fibers in longitudinal view. Moreover, this idea may be used for automatic identification of other types of fibers that have different longitudinal characteristics, for example, cotton and viscose, wool and cashmere.