The eating detection system can track dietary information, aiding in the cultivation of healthy eating habits. However, previous research has primarily focused on detecting eating periods without finely identifying users’ chew count, nor promptly alerting users to insufficient chewing. We propose a real-time dietary detection system called ChewSense, which utilizes earphones to capture intraoral reverse signals, thereby identifying chew counts and the types of food consumed in a single mouthful. ChewSense segments eating signals into chewing segments and validates the authenticity of these segments through pre- and post-action verification. Through a six-month experiment involving 10 volunteers, results demonstrate that ChewSense can accurately identify chew counts and food types, with average accuracies reaching 84.58% and 82.90%, respectively.