At present, in an aquaculture environment, bait casting process of feeding machine mainly uses total time. Such feeding does not consider fish feedingregularity so it wastes bail and causes environmental pollution. High efficiency and intelligent bait casting machine system study thus becomes urgent. An effective way is to set the feeding amount through observing fish aggregation. An intelligent bait casting method based on machine vision technology is proposed in this paper. Through image processing, the geometric area of the fish within the window of the camera and the ratio of the fish area to the widow area were extracted from the pictures, and were taken as characteristic parameters of the pictures. These characteristic parameters can reflect well the crowding level of the fish during the feeding process, and the variation curves of the parameters with time can represent the feeding regularity of the fish. It was also found that the parameter variation curves were similar for different fish. Based on the parameter variation curves, new bait casting scheme was designed. The advantage of using this method is it makes bait casting scientific and reasonable, not only saving feeding cost but also reducing environment pollution.