Green caviar Quality Image Classification based on Enhanced Deeply Learn Model

The quality of Green Caviar products is important in the agricultural and processed food industry in Southeast Asia, especially in Thailand. Due to its characteristics, the quality is difficult to classify, even for the green caviar grower or the consumer. Thus, this research proposes the approach to automatic classification of green caviar quality. The method consists of three main phases. It is firstly, preprocessing. Then, the brunch of green caviar is segmented by causing the input image to have a different background even if it was taken in the control environment. Since its shape is in the oval form, the circular Hough Transform (CHT) is employed to precisely detect its shape. The instances are equally separated into three classes, good quality, average quality, and bad quality, respectively. Secondly, each segmented green caviar is extracted its features i.e., color, circularity, texture etc. The image and features are feed in to a proposed model, ResNet-50, combined with Multi-Layer Perceptron (MLP). Lastly, the result of the classification model is evaluated using confusion matrix compared with another traditional classification model. The proposed model takes 3 hours and 57 minutes in the training step. The result shows that our method reaches 0.934 of overall accuracy. 0.921, 0.944, and 0.938 of accuracy in good, average, and bad quality classes, respectively.