Comparative Analysis of Lightweight Deep Learning Models for Classification of Medical Images

Recently, localization systems tailored for indoor environments have been introduced, leveraging the existing wireless local area network (WLAN) infrastructure.These systems utilize position fingerprinting methods rather than direction-based or time-of-arrival approaches to determine the spatial positions of mobile users.Nevertheless, experimental studies on these systems indicate that factors such as signal attenuation and scattering caused by the higher density of walls significantly impact the accuracy and performance of indoor positioning.Furthermore, the accuracy of indoor positioning systems (IPS) may be compromised by variations in environmental factors, including alterations in height, the introduction or removal of a WLAN Access Point (AP), or modifications to AP power settings.This paper analyses the impact of AP power through a probabilistic analytical model that exclusively utilizes high signal relations to mitigate the effects of low signal relations on WLAN fingerprinting-based IPS, thereby enhancing accuracy performance.A total of 33,300 RSS (Received Signal Strength) data points were collected from five access points and seventy-four reference points to develop the model.The dataset was gathered during the offline learning phase, with RSS readings systematically recorded to maintain consistency.During the online phase, 11,100 data points (33% of the total dataset) were introduced to test the model.Through comprehensive experimental evaluations, the proposed algorithm demonstrates an improvement in the accuracy of the IPS by an average of 15.794% as measured by the Root Mean Square Error (RMSE).The findings suggest that the integration of AP power and high signal relations can substantially enhance the accuracy performance of WLAN fingerprinting-based IPS.