The underwater images of aquaculture ponds, rivers and sea inlets were generally fuzzy and low contrast due to the influence of water turbidity and light attenuation in water. However, the existing literature found that the clarity brought by image enhancement cannot directly improve the detection ability of fish detection model, and even the detection accuracy of the model was degraded. An multiple and outputs blend enhanced method was proposed for fish detection. Blurred underwater images were enhanced by various image enhancement methods, and the enhanced images were input into the fish detection model to obtain multiple outputs. Then the mixed results were postprocessed by non-maximal inhibition method to obtain final test results. Compared with the detection results of the original image, the experimental results on YOLO v3, YOLO v4 tiny and YOLO v4 models showed that the detection accuracy of the proposed method was improved by 2.15 percentage points, 8.35 percentage points and 1.37 percentage points, and the number of fish was increased by 15.5%, 49.8% and 12.7%, respectively. The proposed method achieved the purpose of improving the model detection ability, and it can be applied to fish count and fish category detection.