Rapid Detection of Astaxanthin in Antarctic Krill Meal by Computer Vision Combined with Convolutional Neural Network

To achieve rapid detection of astaxanthin content in Antarctic krill meal, a determination method for astaxanthin content in krill meal was established using computer vision and convolutional neural networks. A total of 70 krill meal samples were analyzed using high-performance liquid chromatography to determine their astaxanthin contents as label, and corresponding images of the samples were acquired using a computer vision system to form the dataset and the dataset was augmented. The model was built using the TensorFlow learning framework. The 5-fold cross-validation was used to tune and evaluate the model and select the optimal parameter model. The optimal parameter model was evaluated by randomly dividing the dataset, and 30 images from the dataset were randomly selected for model validation. The results showed that the optimal hyperparameters model with a root mean square error (RMSE) of 3.59 was preserved through a five-fold cross-validation. For model evaluation, the model was repeated three times. The mean values of the coefficient of determination (R²), mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) for the test set were 0.9626, 1.49, 4.22, and 2.05, respectively. For model validation, the relative errors ranged from 0.10% to 6.46%, indicating small deviations between the predictions and the observations. The astaxanthin content prediction model demonstrated high accuracy, enabling quick and nondestructive detection of astaxanthin content in krill meal samples.