Recognition of animal dog species has always been the focus of the image recognition field, in order to better recognize the canine species in the image and help the society to manage the family pets, this paper discusses to propose a model based on the combination of the YOLOv8 recognition algorithm and the improvement of the network structure of the DenseNet121 for the recognition of canine species, through the addition of the YOLOv8 algorithm to the front of DenseNet121 and the addition of the attention module (CBAM) inside each Dense Block to solve the problems of gradient disappearance, parameter redundancy, and insufficient feature propagation in traditional convolutional network by unique connection. algorithm in front of the DenseNet121 and adding the Attention Module (CBAM) inside each Dense Block, the problems of gradient vanishing, parameter redundancy, and insufficient feature propagation in traditional convolutional networks are solved by a unique connection, which can more accurately recognize and classify the dog in the image. This experiment uses an image dataset containing 120 dog breeds from all over the world from the Kaggle website for image recognition. The cutting-edge deep learning framework pytorch and computationally powerful GPUs were selected to use deep neural networks to train and test the network on dog images, which ultimately improved the accuracy and robustness of the model on image classification and confirmed the reliability of the model.