Monitoring the health and behavior of marine creatures within their environments yields valuable insights into their well-being and responses to different stimuli. Our objective is to employ implantable bio-loggers for the collection and analysis of electrocardiogram (ECG) signals, aiming to uncover the feeding habits of predatory fish. We propose a distinctive processing pipeline tailored for devices with limited resources that can extract high-level data from ECG signals in real-time, including heart rate and feeding events. Our contributions feature practical event recognition algorithms that capitalize on the unique statistical characteristics of heart rate variations linked to feeding. These algorithms reliably detect fish feeding events even within noisy heart rate data. We evaluated our approaches using an in-house logger implanted in twelve coral trout fish, gathering data over a ten-week period. Our signal processing pipeline effectively handles noisy ECG signals. Importantly, while conventional algorithms struggle, our heart rate estimation algorithm achieves an error rate of less than one beat per minute. Additionally, our feeding detection algorithms require considerably fewer computational and energy resources compared to state-of-the-art solutions, all while delivering enhanced accuracy. We implemented our feeding detection and heart rate estimation methods on the bio-logger and assessed the system's overhead.