番茄碰伤和可溶性固形物近红外光谱同时在线检测

可溶性固形物和碰伤是影响番茄品质的两个主要因素。研究的目的是探索可见近红外漫透射光谱同时在线检测番茄碰伤和可溶性固形物的可行性。在单通道送果速度5个每秒条件下，采集番茄近红外漫透射光谱。对比分析碰伤与正常番茄样品的近红外漫透射光谱特性，结果表明，碰伤与正常番茄样品的近红外漫透射光谱在光强上存在明显差异，碰伤果光强要强于正常果，其原因可能是碰伤后果肉变软，透光性变强；在650和675 nm处碰伤果比正常果要多两个吸收峰，可能是碰伤后，番茄表皮颜色发生变化所致。选取贡献率占比最多的前三个主成数，对正常果与碰伤果近红外漫透射光谱主成分定性分析，正常果与碰伤果不能有效聚类，故近红外漫透射光谱主成分定性分析效果不明显，需选择建立高维近红外漫透射光谱定性判别模型。故建立了碰伤番茄样品的近红外漫透射光谱偏最小二乘定性判别模型，误判率为0%，能正确判别碰伤果，故选用碰伤番茄样品的近红外漫透射光谱偏最小二乘定性判别模型作为番茄碰伤果在线剔除分选模型。通过对未参与建模的样品进行验证，能正确识别出碰伤果。经近红外漫透射光谱偏最小二乘定性判别模型剔除碰伤果后，按照可溶性固形物指标进行分级。分别使用全部波段和606～850 nm的波段进行建模预处理，且对全部波段和606～850 nm波段光谱进行2阶导数预处理，前后平滑设为9，利用连续投影算法与遗传算法优选可溶性固形物的光谱建模变量，对比发现，利用未经算法筛选过的606～850 nm波段光谱变量进行建模，效果最好，建立了可溶性固形物在线检测模型，预测集均方根误差为0.43 Brix°。采用未参与建模的样品进行碰伤和可溶性固形物同时在线检测验证，碰伤样品的分选准确率达96%，可溶性固形物样品的分选准确率达91%。表明：番茄碰伤和可溶性固形物近红外漫透射光谱同时在线检测是可行的。

The Online Detection Model Research of Tomatoes’ Bruise and SSD

Soluble solids and bruise are the two main factors affecting tomato quality. The purpose of the study was to explore the feasibility of simultaneous on-line detection of bruises and soluble solids in tomatoes by near-infrared diffuse transmission spectroscopy. The near-infrared diffuse transmission spectra of tomato were collected under the condition of a single-channel fruit delivery speed of 5/s. The near-infrared diffuse transmission spectrum characteristics of the bruised and normal tomato samples were compared and analyzed. The results showed that there was a significant difference in the light intensity between the bruises and the normal tomato samples. The light intensity of the bruises was stronger than that of the normal fruits. The reason may be that the meat becomes soft and the light transmission becomes stronger after the bruising. The two absorption peaks are more than the normal fruit at 650 and 675 nm. The reason may be that the color of the tomato skin changes before and after theinjury. The first three main scores with the highest contribution rate were selected. After qualitative analysis of principal components of near-infrared diffuse transmission spectra of normal fruits and bruises, normal fruits and bruises could not be effectively clustered. Therefore, high-dimensional near-infrared diffuse transmission spectral qualitative discriminant model was selected. By establishing the near-infrared diffuse transmission spectrum partial least squares qualitative discriminant model of the injured tomato sample, the false positive rate of the partial least squares qualitative discriminant model is 0%, which can correctly discriminate the fruit, so the near-infrared diffuse transmission spectroscopy partial least squares qualitative discriminant model of the touched tomato sample was selected as the online knockout sorting model for tomato touch injury. Validation of samples that have not been involved in modeling can correctly identify bruises. After the injurious fruit was removed by the near-infrared diffuse transmission spectroscopic partial least-squares qualitative discriminant model, the classification was based on the soluble solids index. The model is preprocessed using all the bands and the 606～850 nm band, and the second-order derivative preprocessing is performed on all the bands and the 606～850 nm band spectrum, and the front-back smoothing is set to 9, and the continuous projection algorithm and genetic algorithm are used to optimize the soluble solids. The spectral modeling variables, through comparison found that the use of non-algorithm screening 606～850 nm band spectral variables modeling, the best effect, established a soluble solids online detection model, the prediction set root mean square error of 0.43 Brix°. Simultaneous on-line detection of bruising and soluble solids using samples not involved in modeling demonstrated that the accuracy of sorting of bumped specimens was 96%, and the accuracy of sorting of soluble solids samples was 91%. The experimental results show that the simultaneous on-line detection of tomato bruising and soluble solids near-infrared diffuse transmission spectroscopy is feasible.