Sulfur-Fumigated Ginger Identification Method Based on Meta-Learning for Different Devices

Since ginger has characteristics of both food and medicine, it has a significant market demand worldwide. To effectively store ginger and achieve the drying and color enhancement effects required for better sales, it is often subjected to sulfur fumigation. Although sulfur fumigation methods can effectively prevent ginger from becoming moldy, they cause residual sulfur dioxide, harming human health. Traditional sulfur detection methods face disadvantages such as complex operation, high time consumption, and easy consumption. This paper presents a sulfur-fumigated ginger detection method based on natural image recognition. By directly using images from mobile phones, the proposed method achieves non-destructive testing and effectively reduces operational complexity. First, four mobile phones of different brands are used to collect images of sulfur- and non-sulfur-fumigated ginger samples. Then, the images are preprocessed to remove the blank background in the image and a deep neural network is designed to extract features from ginger images. Next, the recognition model is generated based on the features. Finally, meta-learning parameters are introduced to enable the model to learn and adapt to new tasks, thereby improving the adaptability of the model. Thus, the proposed method can adapt to different devices in its real application. The experimental results indicate that the recall rate, F1 score, and AUC-ROC of the four different mobile phones are more than 0.9, and the discrimination accuracy of these phones is above 0.95. Therefore, this method has good predictive ability and excellent practical value for identifying sulfur-fumigated ginger.