Identification and Detection of Freshness in Edible Fishes using IoT and Machine Learning Techniques

Abstract

The identification, and detection of freshness and chemical contaminants in edible fishes are significant aspects of food safety and quality control. The current work is proposed to detect freshness and chemical contamination in fish using IoT and machine learning approaches. The proposed system consists of a formaldehyde sensor that can detect volatile organic compound formaldehyde coated on fish surfaces by the sellers to increase its shelf life. Further, the machine learning model is implemented that can classify the fish samples based on the fish’s iris image of the eye. The real-time dataset is collected from the live fish market. Machine learning models such as Dense Net Algorithm and Efficient Net Algorithm have been used to classify the fresh, and non-fresh fish from the dataset considered by evaluating the model. The EfficientNet and DenseNet Algorithm used have an accuracy of 0.085 and 0.075 respectively for the datasets collected in real time. The model accuracy is evaluated and tabulated. Based on these results, it can be stated that the EfficientNet algorithm has better accuracy than DenseNet. The formaldehyde sensor used has an accuracy of 1-50ppm. The coating of Formaldehyde varies according to the type, variety, and size of the fish. Based on the online references the edible range of formaldehyde content coated on the fish varies from 0-4ppm. If the sensor’s output is less than 4ppm it is considered edible or else it is regarded as unsuitable for consumption. Experimental results show that the proposed system seems to be accurate to identify fresh fish samples and also in detecting formaldehyde content. The proposed work aids the consumers to identify suitable fish for consumption.