The safety of children’s products is closely related to children’s health and life safety, and is an important part of the design process that cannot be ignored. This paper takes children’s products as the research object, and explores the method of children’s product safety assessment from the perspective of image feature recognition and classification. The images of children’s products are preprocessed by geometric transformation, grayscaling and image enhancement to extract their color, texture and shape features. Relying on the image features of children’s products, a VGG-based image feature recognition model of children’s products is constructed, and the model performance is improved by the increase of residual module and the equivalent conversion of multi-branching model into a single-path model, so as to realize the safety assessment of children’s products. In the safety assessment experiments of children’s products, the model in this paper achieves optimal results in four aspects, namely, accuracy, precision, recall and F1 value, which reach 98.34%, 98.34%, 98.33% and 98.34%, respectively. Compared with the original ResNet18 model, the model also has better recognition accuracy in the face of different categories of images of children’s products, and can play an effective role in the work of children’s product safety assessment.
