中文摘要：

假单胞菌是鸡肉腐败最主要的致腐菌,为了快速识别鸡肉中的假单胞菌,首先从腐败鸡肉中分离并筛选出致腐菌,进一步利用聚合酶链反应技术对目标菌株进行生物学鉴别（分别为盖氏假单胞菌、嗜冷假单胞菌、莓实假单胞菌和荧光假单胞菌）;配置鉴定的4种假单胞菌和等体积混合的4种假单胞菌菌液,采集菌液近红外透射光谱信息;然后运用标准正态变量变换对光谱进行预处理,利用联合区间偏最小二乘法筛选出特征波段;最后有比较地运用K最近邻法、最小二乘支持向量机和反向传播人工神经网络建立5种假单胞菌菌液的近红外光谱分类识别模型。其中反向传播人工神经网络模型预测效果最佳,其训练集和预测集的识别率分别为99.17%和95.00%。研究结果表明,近红外光谱结合反向传播人工神经网络可以快速识别鸡肉中的假单胞菌。

英文摘要：

Pseudomonas spp. is the main bacteria involved chicken degradation which ultimately affects the meat quality and the potential of posing health public health threats. The use of near infrared spectroscopy（ NIRS） for rapid identification and monitoring of four strains of Pseudomonas spp. in degrading chicken was attempted. Initially,four Pseudomonas strains namely Pseudomonas gessardii,Pseudomonas psychrophila,Pseudomonas fragi and Pseudomonas fluorescens were isolated from samples of degrading chicken and identified via polymerase chain reaction（ PCR） technology. The different isolated Pseudomonas spp. were cultured in trypticase soy broth（ TSB） and incubated at 30℃ for 12 h to growth.The four isolates of Pseudomonas spp. and their combined mixture in equal proportions were all prepared from the incubated inoculum by using 100 mL∶ 5 mL and each replicated 40 times. The preprocessed data outcomes of the 200 samples using standard normal variable transformation（ SNV） exhibited superiority compared with other deployed data preprocessing algorithms such as multiplicative scatter correction（ MSC）,calibration standard score,first derivative（ DB1） and second derivative（ DB2）. Synergy interval partial least squares（ Si PLS） was employed to select relevant characteristics wavelengths such as 3 999. 64 ~ 4 597. 46 cm~（-1）,6 406. 37 ~ 7 004. 19 cm~（-1）,8 211. 41 ~ 8 805. 38 cm~（-1）and 8 809. 24 ~9 403. 20 cm~（-1）. Principal component analysis（ PCA） was performed prior to the model development with loadings of 97. 02% in PC1,2. 47% in PC2 and 0. 27% in PC3 which indicated the possibility of developing models for the classification of the different samples of Pseudomonas spp. The recognition rates for KNN（ 65. 00%,63. 75%）,SVM（ 91. 67%,86. 25%） and BP-ANN（ 99. 17%,95. 00%） were obtained in the training and prediction sets. The model results obtained for SVM was sufficiently high and may be combined with NIRS system for the possible Pseudomonas spp. classifica