Near infrared (NIR) spectroscopy has been proved to be an effective approach to detect internal quality of agro-products including fruits. Many researches regarding detection of fruits' internal quality are available in literatures, in which one of the most frequently studied is the SSC determination in fruits including apple (Ying et al., 2005), pear (Xu et al., 2012), peach (Golding et al., 2005 and Shao et al., 2011), orange (Jamshidi et al., 2012), watermelon (Jie et al., 2013) and so on. However, because the absorption bands in NIR region are typically broad, weak and extensively overlapped, the appropriate application of NIR spectral data requires careful attention (Arakawa et al., 2011). Furthermore, the modern spectroscopy instrumentations usually possess high resolution, with the spectral data sets obtained often containing hundreds even thousands of variables including noises from environmental and instrumental sources (Wu et al., 2010 and Xu et al., 2012). The large number of spectral variables often leads to complexity and poor predicting ability of a calibration model (Zou et al., 2010). To deal with these problems, chemometrical methods such as partial least squares (PLS) regression and principal component regression (PCR) are commonly used since they are capable to treat very large data matrices and extract relevant information (Leardi and Lupiáñez González, 1998). However, when used for on-line or at-line purposes, the complex calibration models developed with the whole spectrum will not be applicable (Andersen and Bro, 2010). In such cases, multiple linear regression (MLR) models developed with critical variables/wavelengths are practically usable. However, the MLR method suffers from the collinearity among the variables used in the models (Næs and Mevik, 2001). In addition, when building MLR models, variables should be less than samples, which is almost impossible for the large spectral data. Therefore, the selection of effective wavelengths (EWs) is of great significance for building models especially for online purposes (Zou et al., 2010). Elimination of uninformative variables can improve the model predictions, reduce measurement costs and facilitate model interpretation (Andersen and Bro, 2010 and Cai et al., 2008)
ใกล้อินฟราเรด (NIR) (Ying et al., 2005), ลูกแพร์ (Xu et al., 2012), ลูกพีช (ดิงส์ et al., ปี 2005 และ Shao et al. 2011) สีส้ม (Jamshidi et al., 2012) แตงโม (Jie et al., 2013) และอื่น NIR NIR ต้องให้ความสนใจอย่างระมัดระวัง (Arakawa et al. 2011) นอกจากนี้ instrumentations (Wu et al., 2010 และ Xu et al., (Zou et al., chemometrical เช่นสี่เหลี่ยมอย่างน้อยบางส่วน (PLS) (PCR) (Leardi และLupiáñezGonzález, (เซนและ Bro 2010) ในกรณีดังกล่าวหลายถดถอยเชิงเส้น (MLR) / MLR ทนทุกข์ทรมานจาก collinearity ระหว่างตัวแปรที่ใช้ในแบบจำลอง (Naes และ Mevik 2001) อัตราดอกเบี้ย MLR (EWS) (Zou et al., 2010) การกำจัดของตัวแปร uninformative (เซนและ Bro 2010 และ Cai et al., 2008)
การแปล กรุณารอสักครู่..