CAJ | 학술논문

籽棉杂质的分类识别是实现棉花生产线自适应加工的基础与重要依据。该文提出了一种基于局部二值模式和灰度共生矩阵的籽棉杂质分类识别算法，该算法将含杂籽棉图像首先转换为局部二值模式图像，获取图像的微观结构，再用局部二值模式图像生成灰度共生矩阵并计算特征参数，获取图像宏观结构。使用支持向量机作为分类器，用不同尺度的图像结构进行训练，从而达到籽棉杂质的分类识别。试验结果表明，该文设计算法对各种杂质的平均正确识别率达到了94%，超过单独使用局部二值模式和单独使用灰度共生矩阵的正确识别率，为实现棉花自适应加工提供了技术基础。
자면잡질적분류식별시실현면화생산선자괄응가공적기출여중요의거。해문제출료일충기우국부이치모식화회도공생구진적자면잡질분류식별산법，해산법장함잡자면도상수선전환위국부이치모식도상，획취도상적미관결구，재용국부이치모식도상생성회도공생구진병계산특정삼수，획취도상굉관결구。사용지지향량궤작위분류기，용불동척도적도상결구진행훈련，종이체도자면잡질적분류식별。시험결과표명，해문설계산법대각충잡질적평균정학식별솔체도료94%，초과단독사용국부이치모식화단독사용회도공생구진적정학식별솔，위실현면화자괄응가공제공료기술기출。
When cleaning seed cotton, cleaning devices of different types had different cleaning efficiencies on different types of impurities. Therefore, the classification identification of seed cotton impurities had a guiding significance for adjusting the parameter of seed cotton cleaning equipment. A classification recognition algorithm of impurities in seed cotton based on local binary pattern and gray level co-occurrence matrix was proposed in this paper. First, the images were transformed to local binary pattern images, and so the gray value of each pixel was also converted to the local binary pattern value. Local binary pattern reflected the micro-structure of the center pixel and its 3×3 neighborhood, but it could not reflect a wider range of image structure. If the micro-structures of images were similar but macro-structures were different, the local binary pattern could not effectively distinguish the images. Gray level co-occurrence matrix was used to the statistics on the position of pixel pair. The pixel pairs had some relationship of gray values. The distances of pixel pairs could be controlled by the step length. In this paper, gray level co-occurrence matrix was used for local binary pattern image. It could describe the image structures of different scales by adjusting the step-length value. This paper calculated the characteristic values of seed cotton images and all kinds of impurities images with the step-length values from 1 to 8. The characteristics included contrast, angular second moment, correlation and inverse difference moment. The test results showed that these characteristics could distinguish seed cotton and every kind of impurity when the step-length value was equal to 3 or 4. The classifier of this algorithm used the support vector machine. In solving the small-sample, nonlinear and high-dimension problems, the support vector machine had more advantages than the traditional machine learning methods. The support vector machine was a typical two-class classifier. But classification recognition of seed cotton and impurities needed multi-class classifier. Several classifiers of support vector machine were combined into one multi-class classifier, and radial basis function was used as the kernel function of the classifier. This paper compared the standard local binary pattern algorithm (LBP), the standard gray level co-occurrence matrix algorithm (GLCM) and the algorithm designed in this paper (LBP-GLCM). The test results showed that the average recognition rate of the algorithm designed in this paper, which reached 94%, was higher than the LBP algorithm and the GLCM algorithm. Among different objects, the recognition rate of the boll shell and the cotton bush was 100%, the recognition rate of the leaf fragment was 92%, the recognition rate of the dust miscellaneous was 94%, the recognition rates of the deed cotton and barren cotton seed were 90% and 88%, respectively. The recognition rate could satisfy the demand of practical application.