Abstract: The grasshopper detection method based on machine vision was developed for the forecasting and warning of grasshopper disaster. It was used to extract the number of grasshoppers from color videos captured by a locust plague forecasting system with a super-low altitude helicopter. The source image was divided at first into the sky sub-image and the grass sub-image according to the background component. Then the moving zones in the two kinds of sub images were detected respectively by the frame difference method. Finally, the hopped grasshoppers were recognized by classifying the found moving zone with the shape feature of grasshoppers. The automatic detection number of hopped grasshoppers were introduced into the number prediction model built by the connection of hopped grasshoppers and ground grasshoppers. Therefore, the number of ground grasshoppers could be computed indirectly. The experiment results showed that the recognition rate of hopped grasshoppers was 80%～100%, and the precision of the number of ground grasshopper computed by the built mathematic model achieved 80%. The grasshopper detection method based on machine vision can satisfy with the demand of precision grasshopper prediction.