基于可见近红外光谱及增强回归树算法的鸡蛋种类鉴别
Identifying egg varieties based on boosting regression trees algorithm and visible near infrared spectrum

利用可见-近红外光谱技术，选取湖北地区同一品种不同饲养环境下的鸡蛋，提取鸡蛋的光谱透射率（500～900 nm），利用标准正态变量变换对光谱数据进行预处理，结合竞争性自适应重加权与主成分分析对光谱数据进行二次降维，并将提取的特征信息输入增强回归树算法，建立鸡蛋土洋种类鉴别模型，模型的训练集和测试集判别正确率分别为98.33%和97.00%。结果表明，应用基于可见 近红外光谱及增强回归树方法，针对同一母鸡品种但不同饲料产出的土洋鸡蛋的种类鉴别是可行的。
The varieties of soil and native eggs relate to its internal quality and sales price. Identifying the egg types quickly and nondestructively will be of great significance to regulate the market of agricultural products. The visible/near infrared spectrum technology was used to extract the spectral transmittance (500-900 nm) of free-range and ordinary of the same egg variety collected from different breeding environment of Hubei Province. The spectral data were pretreated by the standard normal variate (SNV).The competitive adaptive reweighed sampling(CARS) combined with the principal components analysis(PCA) method was used to perform two times dimensionality reduction of spectral data. The processed data were transmitted as the input of boosting regression trees(BRT) and established the model for identifying egg varieties. The correct rate of the model set and the prediction set are 98.33% and 97.00%. The results showed that applying visible near infrared spectrum based on boosting regression trees to identify eggs with the same hen breeds but different feeds is feasible.