CAJ | 학술논문

针对全波段光谱技术的生鲜猪肉综合品质快速无损分类存在光谱数据量大、样本数量较少时分类准确率较低等缺点。该文提出了一种基于偏最小二乘（partial least squares，PLS）投影分析算法和支持向量机的生鲜猪肉综合品质分类器。利用基于偏最小二乘投影分析算法对全波段光谱数据进行数据降维，选取了13个特征波长。利用粒子群优化算法优化支持向量机惩罚参数和径向基核函数参数，优化后二者最优为4.939和0.01。利用选取的特征波长和优化后的参数建立了生鲜猪肉综合品质支持向量分类器。研究结果表明，分类器对训练集中白肌肉（pale, soft and exudative，PSE）、正常肉（reddish-pink, firm and non-exudative, RFN）和黑干肉（dark, firm and dry, DFD）的回判识别率分别为为88.46%、94.11%和92.31%；测试集中PSE、RFN和DFD预测正确率分别为84.62%、94.11%和84.62%。该分类器满足模型简单、预测准确率高等优点，为生鲜猪肉综合品质在线分级提供参考。
침대전파단광보기술적생선저육종합품질쾌속무손분류존재광보수거량대、양본수량교소시분류준학솔교저등결점。해문제출료일충기우편최소이승（partial least squares，PLS）투영분석산법화지지향량궤적생선저육종합품질분류기。이용기우편최소이승투영분석산법대전파단광보수거진행수거강유，선취료13개특정파장。이용입자군우화산법우화지지향량궤징벌삼수화경향기핵함수삼수，우화후이자최우위4.939화0.01。이용선취적특정파장화우화후적삼수건립료생선저육종합품질지지향량분류기。연구결과표명，분류기대훈련집중백기육（pale, soft and exudative，PSE）、정상육（reddish-pink, firm and non-exudative, RFN）화흑간육（dark, firm and dry, DFD）적회판식별솔분별위위88.46%、94.11%화92.31%；측시집중PSE、RFN화DFD예측정학솔분별위84.62%、94.11%화84.62%。해분류기만족모형간단、예측준학솔고등우점，위생선저육종합품질재선분급제공삼고。
Pale, soft, and exudative (PSE) and dark, firm, and dry (DFD) are degraded grades of pork. Reddish-pink, firm, and non-exudative (RFN) are considered superior grades of pork. The meat color, pH value, and water holding capacity for PSE, DFD, and RFN have different ranges which directly influence the purchasing decision of customers, thus affecting the economic benefits of the pork processors. With growing demand for quality meat and increasing state-of the-art meat processing technologies, pork industries are in need of a reliable technology for rapid, accurate, and non-destructive detection of meat quality. Spectral technology has gained importance in agricultural research and the meat industry. Spectral technology has also shown its massive potential in the meat industry. However, full wave band contains huge amount of spectral information, which possess a severe drawback to spectroscopic technology in terms of its accuracy and detection speed. This study proposes a pork comprehensive quality classifying method based on partial least squares (PLS) projection algorithm and support vector machine (SVM). The acquired spectral data of sample meat were first normalized using standard normal variation (SNV) transformation method. Next, the whole sample was randomly classified into two sets: the calibration set (for developing prediction model using 75%of total samples) and the validation set ( to validate the model using remaining 25%of the total samples). Mean filter was used to smooth the normalized spectral signal. Partial least square projection method was then used to obtain projection coefficients for each wave band. Classification models were developed using 1 to 20 different wave bands, and root mean square errors (RMSE) of each model were calculated by cross validation. The lowest RMSE was obtained using 13 wave bands and it was observed that it became stable with more wavelengths. The spectral wave lengths of 371, 388, 425, 456, 473, 562, 578, 607, 696, 764, 772, 813 and 927 nm were chosen based on the RMSE value to develop a pork quality prediction and classification model. Particle swarm optimization (PSO) algorithm optimized penalty parameter and radial basis core function parameter, was 4.939 and 0.01 for the best. Pork comprehensive quality SVM classifier was established using special wavelengths and optimized parameters. It was shown that the back-to-recognition rate of pale, soft, and exudative (PSE), reddish, firm, and non-exudative (RFN), and dark, firm, and dry (DFD) in training sets were 88.46%, 94.11%, and 92.31%respectively. For testing sets, predictive accuracy rates of three kinds were 84.62%, 94.11%, and 84.62%respectively. The study revealed the advantages of the established classifier, such as based on simple model and achieving high prediction accuracy, and so on. This study shows a simple, rapid, and non-destructive method by optical instrument to detect and classify pork based on its comprehensive quality. The study can be a milestone to develop a fast, accurate, and reliable technique for non-destructive detection of pork in slaughtering house, super markets and other required areas where quality is a concern.