Abstract: Red claw crayfish has been one of the most popular aquatic foods in recent years. It is of great significance to rapidly and accurately detect the typical feature information during artificial breeding, such as the position, size, and color of densely distributed shrimp eggs. The breeding quantity, time, and quality indicators can be indirectly estimated from these feature information. Conventional image processing can usually rely on manually setting features and thresholds, leading to limited applicability and accuracy in complex practical conditions. Thus, there is a high demand for the practical scenes of highly dense shrimp eggs. Deep learning can be used to automatically learn the complex patterns and features in large datasets, in order to improve detection accuracy and generalization. In this study, a new detection was proposed for the typical features of densely distributed shrimp eggs using deep learning. Firstly, the images of shrimp eggs in the dataset were captured at the Balidian aquaculture base of the Zhejiang Institute of Freshwater Fisheries. The dataset contained 450 manually annotated shrimp egg images, which were a total of about 260000 accurately labeled shrimp eggs. A 7:3 ratio of the image dataset was taken as the training and validation datasets for the model training and the performance test. Then, several typical regression models were used to estimate the density map of shrimp eggs images. The local peak filtering was applied on the estimated density maps, in order to locate all the shrimp eggs in the image. A comparison was made on the density map quality and location accuracy of different regression models. Meanwhile, the RECNet using the residual module network achieved the highest quality of density map and a location accuracy of 94%. Next, according to the location points of the shrimp eggs, the K-dimension tree was used to calculate the average distance between the target point and the three nearest adjacent location points, in order to generate an anchor box for each shrimp egg. Furthermore, a simple boundary search was used to obtain a tight bounding box of the shrimp egg, due to the too-large anchor boxes of the shrimp eggs in sparse regions. Then a subimage surrounding a single shrimp egg was cut from the entire image. The elliptical contour detection was used to obtain the size of the single shrimp egg in the sub image. Compared with the measurement size, the errors of size in the long and short axes of the shrimp egg were 0～2 pixels, and the minimum and maximum errors of the area of shrimp eggs were 4 and 11 pixels, respectively. At last, the K-means color clustering was used to detect the color composition of densely distributed shrimp eggs images and their subimage. The shrimp eggs image was clustered with K values to obtain the K clusters, namely K main colors. Similarly, the color of a single shrimp egg after the elliptical contour detection was clustered into two main colors to represent the egg and the transparent part of the egg, together with the remaining background color. The color detection can potentially provide important data to evaluate the shrimp egg maturity. The finding can also provide a strong reference in shrimp egg breeding, indicating the promising practical application. The location, size, and color detection can offer the basic supporting data for the evaluation of basic indicators and the optimization of incubation management in the artificial incubation of red claw crayfish.