Joint recognition for location and activity based on multidimensional features of CSI images

The joint recognition technology of location and activity, leveraging CSI (Channel State Information), finds widespread applications in domains such as human-computer interaction and smart homes, owing to the pervasive deployment of wireless networks. While single-dimensional recognition algorithms have achieved notable accuracy, there remains room for improvement in multidimensional recognition algorithms. To address this gap, a joint recognition algorithm for location and activity, termed the JRLA-MFCI algorithm, is proposed based on multidimensional features extracted from CSI images. This algorithm utilizes fine-grained CSI information to construct anti-interference CSI amplitude difference and phase difference images. Texture and color features are subsequently extracted to form feature vectors, and the SVM algorithm employing a quadratic polynomial kernel function with optimized parameters is employed for the joint recognition of location and activity. Extensive experimentation validates the efficacy of the JRLA-MFCI algorithm, yielding recognition accuracies of 95.86% and 93.86% in two distinct experimental scenarios, respectively.