Identification of oil authenticity and adulteration using deep long short-term memory-based neural network with seagull optimization algorithm

One of the most important aspects of people's everyday diet is edible oils. Good quality cooking oil plays a key role in one's health. Due to the increased demand for oil in both the international and domestic markets, vendors often mix the high-quality oil with low-quality ones causing adulteration which is a serious issue to be solved. Thus, qualified (authentic or pure) edible oils are expensive. Gall bladder cancer is mainly caused when the oil is adulterated with butter yellow, argemone oil, mixing good quality oil with low-quality oils, and wrong ingredients with fraudulent labeling. In the past decades, spectrophotometric methods and machine learning techniques are utilized for adulteration and authenticity identification of sunflower oil, olive oil, corn oil, coconut oil, mustard oil, soybean oils. Nevertheless, the performance of these methods is decreased due to data imbalance, overfitting, higher cost, more execution time, computational complexity, and inaccurate classification. To tackle these issues, we have proposed Deep Long Short-Term Memory (LSTM) neural network with a Seagull Optimization Algorithm (SOA) for the authenticity and adulteration of edible oils classification. In this study, 5 kinds of edible oils such as coconut oil, rice oil, sesame oil, sunflower oil, and Olive oil are used. Each of the oil samples was kept in the refrigerator at 4 degrees C. During data acquisition, the proton resonance frequency was 19.91 MHz and the magnetic field strength was 0.467 T. The obtained signals are applied for edible oil classification, which is handled using a deep LSTM neural network with SOA. Based on the experimental investigation, the proposed method accomplished superior performances than existing methods including LFNMR-CNN, LFNMR-SVM, DLC, Pre-trained CNN.