High-Precision Indoor Positioning for Drone Based on EfficientNetV2 and CNN

With the development of smart hardware devices and the increasing demand for indoor positioning, Wi-Fi Re-ceived Signal Strength Indicator (RSSI)-based fingerprinting has become a simple and low-hardware requirement method for indoor positioning. However, in complex indoor environments, signal attenuation and multipath effects can severely interfere with wireless signal propagation, thereby reducing positioning accuracy. This paper proposes a new positioning scheme that uses an EfficientNetV2 classification model and a Convolutional Neural Network (CNN) regression model. The scheme first converts RSSI values into feature maps, then uses the EfficientNetV2 model to classify the wall where the drone is located, narrowing down the positioning area. Then the feature maps are input into the proposed CNN model to estimate three-dimensional coordinates and improve positioning accuracy. Experimental results show that the proposed method achieves an average error of 2.01 meters in three-dimensional positioning Euclidean distance, compared to 2.14 meters for the traditional method, resulting in a 6.07 % improvement in positioning accuracy. These results demonstrate the advantages of the new method in overcoming signal interference in indoor environments and effectively improving positioning accuracy.