As a typical vertebrate model animal, zebrafish has been widely studied in the biomedical field because of its high similarity with human genes and transparent body for convenient observation. It is vital to recognize and monitor the heart of zebrafish larvae to study the pathogenesis of cardiovascular diseases and drug effects. A high-throughput heart recognition and monitoring method based on micro-vision was proposed to solve the problems of micromanipulation in the manipulation and observation of anisotropic zebrafish larvae. First, the Gaussian mixture model was used to determine the positions and directions of larvae. After one of the larvae was moved to the center of the POV of the microscope, a convolutional neural network was built to identify the roll angle of the larvae, and the larvae in abnormal postures were adjusted to make their hearts easy to monitor. After increasing magnification, a heart monitoring algorithm based on pixel intensity was proposed to realize heartbeat monitor. Experimental results show that the proposed system could realize accurate recognition and provide graphs of heartbeats and beat-rates of zebrafish larvae, which provides a novelty, stable, and efficient experimental observation method for the biomedical research of zebrafish.