Classification of Impurities in Machine-Harvested Seed Cotton Using Hyperspectral Imaging
CHANG Jin-qiang, ZHANG Ruo-yu*, PANG Yu-jie, ZHANG Meng-yun, ZHA Ya
College of Mechanical and Electrical Engineering, Shihezi University/Key Laboratory of Northwest Agricultural Equipment, Ministry of Agriculture, Shihezi 832003,China
Abstract:The classification and detection of impurities in machine-harvested seed cotton provides a reference for adjusting cotton cleaning mechanical processing parameters and has important significance for improving lint quality. However, the uneven distribution of seed cotton makes image detection more difficult, and traditional detection methods cannot effectively detect various impurities. The hyperspectral imaging method was used to discriminate the five impurities (cotton leaf, cotton stem, plastic film, hull inner, and hull outer) in the machine-harvested seed cotton. The hyperspectral images of 120 machine-harvested seed cotton samples were collected, and the region of interest was selected to obtain the average spectral curve. Due to the difference in the composition of materials, various impurities showed different spectral absorption and reflection characteristics, and the spectral difference of the characteristics of different materials was greater than that of similar materials. Principal component analysis (PCA) of the extracted average spectral curve showed that cotton, plastic film and hull outer were better separable than the other three types. However, the spectral distributions of cotton leaf, hull inner, and cotton stem overlapped seriously. Based on the extracted average spectral curve as the training sample, three discrimination algorithms of linear discriminant analysis (LDA), support vector machine (SVM) and neural network (ANN) were used to optimize the algorithm parameters and finally established the impurity detection model. Among them, the sample space after dimensionality reduction of the LDA model shows better separability than PCA. Regularization was used to prevent overfitting in LDA, and the accuracy rate of the training set was 86.4%, and the accuracy of the test set was 86.2%. The parameter optimization result of the SVM model was C=105, g=0.1. The accuracy of the training set was 83.42%, and the accuracy of the test set was 83.40%. The parameter optimization result of the ANN model was that the number of hidden layers and neurons were 1 and 10, respectively. The accuracy rate of the training set was 82.9%, and the accuracy rate of the test set was 81.8%. Comparing the classification accuracy and detection time of the three models, the results of the LDA model were all the best. Through the pixel level discrimination of hyperspectral images, the results show that both cotton and botanical impurities were effectively detected. However, there were misidentifications between plastic film and cotton, which was consistent with the results of the impurity spectrum classification discrimination model. Therefore, hyperspectral imaging technology can detect and identify seed cotton impurities quickly and non-destructively and provide feedback parameters for cotton processing equipment, which is of great significance to the mechanization and intelligence of cotton processing.
[1] WANG Xin, LI Dao-liang, YANG Wen-zhu, et al(王 欣, 李道亮, 杨文柱, 等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2015, 46(8): 7.
[2] Zhou W H, Xu S D, Liu C J, et al. Applied Spectroscopy Reviews, 2016, 51(4): 298.
[3] ZHANG Zhi-feng, ZHAI Yu-sheng, GUO Ying-ying, et al(张志峰,翟玉生, 郭莹莹,等). Laser & Optoelectronics Progress(激光与光电子学进展), 2015, 52(3): 154.
[4] ZHANG Lin, WEI Ping, WU Jian-bo, et al(张 林, 韦 平, 伍剑波, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016, 32(15): 289.
[5] ZHANG Cheng-liang, LI Lei, DONG Quan-cheng, et al(张成梁, 李 蕾, 董全成, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016, 32(24): 189.
[6] ZHANG Cheng-liang, LI Lei, DONG Quan-cheng, et al(张成梁, 李 蕾, 董全成, 等). Transactions of the Chinese Society of Agricultural Machinery(农业机械学报), 2016, 47(7): 28.
[7] WANG Hao-peng, LI Hui(王昊鹏, 李 慧). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2015, 31(3): 236.
[8] Ni Chao, Li Zhenye, Zhang Xiong, et al. IEEE Access, 2020, 8: 93028.
[9] Wang Xin, Yang Wenzhu, Li Zhenbo. Computers and Electrical Engineering, 2015, 46: 500.
[10] Fortier C, Cintron M S, Rodgers J. AATCC Journal of Research, 2015, 2(6): 1.
[11] Jiang Yu, Li Changying. PLOS ONE, 2015, 10(3): e0121969.
[12] Jiang Y, Li C. Computers and Electronics in Agriculture, 2015, 119: 191.
[13] Mustafic A, Li C. Textile Research Journal, 2015, 85(12): 1209.
[14] Zhang R Y, Li C Y, Zhang M Y, et al. Computers and Electronics in Agriculture, 2016, 127: 260.
[15] Zhang M Y, Li C Y, Yang F Z. Computers and Electronics in Agriculture, 2017, 139: 75.