Abstract: With the use of many ground-based and space-based solar telescopes, a large amount of solar image data has been generated. The study of solar active region measurement based on photosphere image is an important topic. For example, the problems of sunspot formation and evolution, sunspot rotation and power transmission require a lot of statistical analysis of sunspots. Therefore, the automatic identification of sunspots is a key step to carry out research. In this paper, according to the morphological characteristics of sunspots and their positions on the solar plane, sunspots were divided into 6 categories: regular sunspots, irregular sunspots, small sunspots, sunspot groups, edge sunspots and edge sunspot groups and a sunspot recognition model,VanillaNet-MBConv-YOLOv8 was proposed. The automatic recognition of these 6 types of sunspots in full-day photosphere image was realized. The network proposed replaced the original Darknet53 backbone network of YOLOv8 with VanillaNet, optimized the original C2f module with MBConv module, thereby enhancing the model's ability to extract sunspot features, improving the phenomenon of error reduction and missing detection of the network model, and increasing the detection accuracy. The experimental results on SDO/HMI(solar dynamics observatory/helioseismic and magnetic imager) full-day photospheric images show that compared with the original YOLOv8 model and other mainstream network models, the proposed network model has significantly improved network performance, recognition accuracy and missing detection phenomena. The precision could reach 91.0%, recall could reach 71.5%, and F1-score could reach 0.80.