In aquaculture, the growth of fish pathogenic bacteria can cause outbreaks of bacterial fish diseases and trigger the death of a large number of cultured fish. Therefore, it is extremely important to monitor the concentration of bacterial fish disease pathogenic bacteria. To count the number and concentration of small targets quickly and accurately such as fish disease pathogenic bacteria, deep learning was introduced into aquaculture, and an improved detection algorithm for fish disease pathogenic bacteria based on YOLOv5 was proposed. First, a layer of bottom-up path enhancement is added to the path aggregation network structure, and feature fusion is performed with the feature map output from the first CSP module in the backbone feature extraction network to improve the accuracy of fish pathogenic bacteria detection. Then the attention mechanism is added after each convolutional module in the backbone feature extracting network to further refine the features extracted by the convolutional module. Finally, the K-means clustering algorithm is used for the fish pathogenic bacteria dataset to obtain a priori frames that match more closely with the feature maps. It is shown that the mean average accuracy of the improved algorithm on the test set is 69.19%, an improvement of 2.34%, It is verified that adding an upsampling layer and an attention mechanism has a good effect on the detection of small targets such as fish pathogenic bacteria. It is shown that the improved algorithm in this paper can accurately detect bacteria, effectively improving the detection accuracy of fish-pathogenic bacteria and reducing the rate of missed and wrong detection. The method can also be applied to the detection and identification of fish and shrimp eggs by supplying feed and oxygen matched to the number of eggs, etc.