Secured CNN Indoor Localization Resisting Spoofing

The patent applies preliminary action by generating a second training data set before deployment using statistical distribution models of radio signal strengths. This pre-processing step creates augmented training data that accounts for potential attack scenarios, enabling the neural network to learn resilient features in advance rather than reacting to attacks during operation.

The patent changes parameters by applying random or temporally correlated offset values to the training data, simulating various attack conditions. This parameter transformation allows the neural network to be trained on diverse signal strength variations, improving its ability to distinguish between legitimate signal changes and malicious attacks.

The patent uses copying by creating a second training data set that replicates the structure and characteristics of the first training data set, but with augmented variations. This copying approach with statistical modeling allows comprehensive attack scenario simulation without requiring entirely new data collection processes.

The system applies self-service by using the statistical distribution models derived from the first training data set to automatically generate the second training data set. The neural network training process itself produces the statistical parameters needed for data augmentation, eliminating the need for external intervention or manual attack scenario creation.