This study develops a reliable radial basis function neural networks (RBFNNs) to estimate freshness for tilapia fillets stored under non-isothermal conditions by using optimal wavelengths from hyperspectral imaging (HSI). The results show that, for tilapia fillet stored at −3, 0, 4, 10, and 15 °C and non-isothermal conditions, total volatile basic nitrogen (TVB-N), total aerobic counts (TAC), and the K value increase whereas sensory scores decrease with increasing storage time. To simplify the models, nine optimal wavelengths were selected by using the successive projections algorithm (SPA), following which SPA–RBFNN models were built based on the selected wavelengths and the values of TVB-N, TAC, K, and sensory evaluations for tilapia fillets store isothermally. The ability of the models based on HSI to predict the freshness indicators were verified for tilapia fillets stored under non-isothermal conditions. HSI thus has an excellent potential for nondestructive determination of freshness in tilapia fillets.

这项研究开发了一种可靠的径向基函数神经网络（RBFNN），通过使用高光谱成像（HSI）的最佳波长来估计在非等温条件下存储的罗非鱼片的新鲜度。结果表明，对于在-3、0、4、10和15°C和非等温条件下存储的罗非鱼片，总挥发性碱性氮（TVB-N），总需氧量（TAC）和K值随着储存时间的增加，感官分数降低。为了简化模型，使用连续投影算法（SPA）选择了9个最佳波长，然后根据所选波长和TVB-N，TAC，K值构建SPA–RBFNN模型和罗非鱼片的感官评估等温保存。针对非等温条件下存储的罗非鱼片，验证了基于HSI的模型预测新鲜度指标的能力。因此，HSI在无损确定罗非鱼片中的新鲜度方面具有极好的潜力。