Matched Filter Interpretation of CNN Classifiers with Application to HAR

Abstract

Time series classification is an active research topic due to its wide range of applications and the proliferation of sensory data. Convolutional neural networks (CNNs) are ubiquitous in modern machine learning (ML) models. In this work, we present a matched filter (MF) interpretation of CNN classifiers accompanied by an experimental proof of concept using a carefully developed synthetic dataset. We exploit this interpretation to develop an MF CNN model for time series classification comprising a stack of a Conv1D layer followed by a GlobalMaxPooling layer acting as a typical MF for automated feature extraction and a fully connected layer with softmax activation for computing class probabilities. The presented interpretation enables developing superlight highly accurate classifier models that meet the tight requirements of edge inference. Edge inference is emerging research that addresses the latency, availability, privacy, and connectivity concerns of the commonly deployed cloud inference. The MF-based CNN model has been applied to the sensor-based human activity recognition (HAR) problem due to its significant importance in a broad range of applications. The UCI-HAR, WISDM-AR, and MotionSense datasets are used for model training and testing. The proposed classifier is tested and benchmarked on an android smartphone with average accuracy and F1 scores of 98% and 97%, respectively, which outperforms state-of-the-art HAR methods in terms of classification accuracy and run-time performance. The proposed model size is less than 150 KB, and the average inference time is less than 1 ms. The presented interpretation helps develop a better understanding of CNN operation and decision mechanisms. The proposed model is distinguished from related work by jointly featuring interpretability, high accuracy, and low computational cost, enabling its ready deployment on a wide set of mobile devices for a broad range of applications.

A Tiny Matched Filter-Based CNN for Inter-Patient ECG Classification and Arrhythmia Detection at the Edge

“…In Ref. [ 7 ], we presented the MF interpretation of CNN classifiers with application to human activity recognition. The developed model achieves superb classification performance with significantly reduced complexity compared to related models.…”

“…Despite their prevalence, CNNs are employed as “black box” models because their internal operations and decision mechanisms are not explicitly understood [ 8 ]. The MF theory has been exploited to explain the CNN operation in the time domain [ 7 ]. An MF is an optimal filter for signal detection in the presence of noise.…”

“…Moreover, the inter-patient classification of ECG signals is a challenging ML problem because the training and testing sets come from different sources with inherited inter-individual variations. In response, the matched filter (MF) interpretation of the convolutional neural network (CNN) presented in our previous work [ 7 ] is exploited to address these challenges and develop a tiny ECG classifier for edge inference. MFs are optimal filters for signal detection in the presence of noise, and their operation is tightly linked to CNNs [ 8 ].…”

Design and Analysis of Convolutional Neural Layers: A Signal Processing Perspective

“…Some works provide interpretations and implementations of DSP algorithms using full DNN and CNN architectures [43,44]. In our previous works, we presented MFand STFT-based CNN classifiers for ECG classification and human activity recognition (HAR) [45,46,47]. The proposed models achieved remarkable classification and real-time performance results compared to state-of-the-art rivals at a much lower computational cost.…”

“…Potential applications of the proposed method include health monitoring [46,47], machine-human interface [48], activity recognition [45], machinery fault diagnosis [19,17,18], audio signal processing [27,20,10], radar signal processing [36], automatic modulation recognition [49,50], wireless sensor networks and internet of things [51,52], and communication systems [38,37], to name a few. In these applications, data is collected as a time series and fed to an ML model to perform a specific task such as classification, regression, or forecasting.…”