In aquaculture, information about the fish appetite would be a valuable input into the process of developing efficient feeding management strategies, it holds important information for aquaculturist. In recent years, according to the fish behavior, automatic and objective assessment of their appetite is the future development trend. In order to achieve an objective and accurate assessment of fish appetite, this study proposes a method for quantifying fish appetite based on near-infrared machine vision. Specific objectives of this study were to: 1) solve the issue of low contrast and light reflection caused by low and varying illumination, 2) develop an algorithm to extract index that can assess and quantify the flocking level of fish in near infrared images, 3) design an algorithm to extract index that can assess and quantify the snatch strength of fish in near infrared images. The specific implementation process of this study is: 1) After using the support vector machine to classify the reflective frames, the contrast of the image was enhanced by the Multi-Scale Retinex algorithm and the gray scale nonlinear transformation, 2) The flocking index of fish feeding behavior (FIFFB) was extracted by Delaunay Triangulation. 3) The snatch intensity of fish feeding behavior (SIFFB) was extracted by image texture and gray level gradient co-occurrence matrix (GLGCM). And the performance of the method was also evaluated. The results showed that appetite changes during fish feeding can be accurately assessed by FIFFB and SIFFB. Among them, the linear correlation index of FIFFB and artificial expert scoring results can reach 0.95, and correlation coefficient of SIFFB and area method can up to 0.89. Finally, the real-time description of fish feeding process were realized, which lays a theoretical foundation for developing fine feeding equipment and guiding practice.