Aiming at the the problem of complex deployment, poor robustness, and low positioning accuracy in visible light indoor positioning systems based on received signal strength (RSS), a visible light indoor positioning perception model based on convolutional neural network(CNN) visual imaging is proposed, and the layout method of light sources is studied. Firstly, a visible light visual imaging positioning perception model is designed and built based on ambient light and ordinary light-emitting diode(LED) light sources. Then, the deep features of the image are extracted through the pre-trained CNN model. Based on this,the positioning accuracy model of the indoor positioning perception model based on CNN visual imaging is optimized by studying the relationship between positioning accuracy and the number of light sources, as well as the spacing between light sources indifferent light source layout models. The experimental results show that compared with the indoor positioning perception model based on RSS, when the positioning errors are less than 2.1 cm and 3.9 cm respectively, the confidence probabilities of the proposed model are increased by 10% and 6.7% respectively. At the same time, compared with the rectangular layout method and the triangular layout method, the positioning accuracy of the cross layout method is improved by 9.5% and 16% respectively.