Indoor location perception model based on Resnet50 and Elman network

The visible light indoor position perception method not only solves the limitations of traditional positioning technology indoors, but also promotes innovation in fields such as smart retail and healthcare with its advantages of high accuracy and low cost. At present, visible light indoor positioning methods based on received signal strength are restricted by issues such as environmental interference and poor signal stability. Additionally, traditional feature extraction methods result in insufficient diversity in feature databases, and the initial parameters of positioning models tend to fall into local optima, leading to significant fluctuations in positioning errors in complex indoor scenarios and making it difficult to consistently achieve high-precision positioning results. Therefore, it is necessary to address the problems of weak signal anti-interference capability, inadequate feature representation, and insufficient model parameter optimization by focusing on three core aspects: data collection, feature extraction, and model optimization, thereby providing a technical pathway to enhance the accuracy and stability of indoor positioning. In this regard, to improve the stability and accuracy of this technology, the design of light source layout and image data acquisition method was studied. This research developed an image data feature extraction method based on Resnet50 and referenced the idea of feature pyramid. Meanwhile, an indoor location sensing model based on Elman network was designed, and an improved grey wolf optimization algorithm was proposed to optimize the initial weights and threshold parameters of the Elman network. The results showed that the average cosine similaritycorresponding to the feature extraction methoddesigned in the studywas 0.103, which wascloser to 0, indicating that the feature library constructed by this method was more diverse. Under different test functions, the average fitness values of the improved grey wolf optimization algorithm were 2.69 × 105, 1.47 × 101, and 2.17 × 102, respectively, all of which were lower than the comparison algorithm. At different heights, the average errors of the designed perception model were 3.04 cm, 3.57 cm, and 3.19 cm, respectively, which were notably lower than the comparison models. The design model also performed better in dynamic analysis and environmental light impact analysis. The designed perception model exhibited excellent performance and is capable of offering technical assistance for improving indoor positioning accuracy.