新鲜度是即食海参加工品质调控和贮藏品质监控的关键指标。 针对感官评定和现有理化检测无法满足即食海参产品大批量、 标准化、 工业化生产问题, 提出了一种基于高光谱图像的即食海参新鲜度快速无损检测方法, 通过图像主成分分析和波段比运算相结合, 优选特征波长和图像; 依据海参腐败机理, 建立图像纹理特征与即食海参新鲜度等级间的关联模型, 实现即食海参新鲜度无损、 快速评价。 首先针对高光谱图像巨大的数据量展开降维研究。 根据即食海参体壁光谱吸收特性, 以具有明显化学吸收特征的波长(474和985 nm)为分界点, 获得包括全检测波段(400～1 000 nm)在内的六个待处理波段, 通过分段图像主成分分析实现待测波段的优选, 利用权重系数和波段比图像运算, 最终将686和985 nm波段比图像确定为特征图像。 面向特征图像的感兴趣区域(ROI), 构建灰度共生矩阵(gray-level co-occurrence matrix, GLCM)、 灰度梯度共生矩阵(gray-gradient co-occurrence matrix, GGCM)、 改进的局部二元模式纹理描述子(local binary pattern, LBP), 分别提取纹理参数作为输入, 以挥发性盐基氮(total volatile basic nitrogen, TVB-N)检测为标准, 建立经粒子群优化的BP 神经网络(back propagation, BP)即食海参新鲜度判别模型, 新鲜度等级判别准确率分别为90%, 95%和80%。 结果表明, 即食海参高光谱图像灰度梯度共生矩阵的纹理特征用于新鲜度判别效果较好。 为即食海参新鲜度快速无损检测方法研究和仪器开发提供了理论基础和数据支持。

Freshness is a key index for quality regulation and assurance during the processing and storage of ready-to-eat sea cucumbers. The usual freshness detection methods, sensory evaluation and physicochemical detection, are inadequate for mass standardized and industrial production. In this study, a nondestructive freshness detection method based on hyperspectral imaging was proposed for ready-to-eat sea cucumber (RTESC). The characteristic wavelengths and images were first selected using Principal Component Analysis (PCA) and band ratio algorithm. According to the rottenness mechanism of RTESC, the correlation model between the texture features of hyperspectral images and the freshness degree of RTESC was established to achieve a fast, non-destructive and non-invasive evaluation of RTESC freshness. The effective dimensional-reduction method was adopted to address the massive data of hyperspectral images. According to the spectral absorption characteristics of the sea cucumber body wall, the wavelengths (474 and 985 nm) with significant chemical absorption characteristics were used as dividing points for band division. Thus, five sub-bands and the full band (400～1 000 nm) were acquired for data processing. Next, the bands were optimized using Image Principle Component Analysis (IPCA). Based on the calculated weight coefficients, the band-ratio image at 686 and 985 nm was selected as the characteristic image. On that basis, the gray-gradient co-occurrence matrix (GGCM), gray-level co-occurrence matrix (GLCM), and local binary pattern (LBP) descriptor were constructed to extract texture features. Meanwhile, the measured total volatile basic nitrogen (TVB-N) contents were used as the criterion. Using these three types of texture features as the input data, three freshness evaluation models based on particle swarm optimization (PSO) and back propagation neural network (BPNN) were established. The detection accuracies of these three models are 90%, 95%, and 80%, respectively. The results show that, using the texture characteristics extracted by GGCM from the hyperspectral images, the detection performances are favorable. The present study provides theoretical foundations and technological supports for the development of fast and non-destructive detection methods for RTESC.