20 January 2020 in IEEE Communications Letters
This paper proposes a cost-efficient convolutional neural network (CNN) for robust automatic modulation classification (AMC) deployed for cognitive radio services of modern communication systems. The network architecture is designed with several specific convolutional blocks to concurrently learn the spatiotemporal signal correlations via different asymmetric convolution kernels. Additionally, these blocks are associated with skip connections to preserve more initially residual information at multi-scale feature maps and prevent the vanishing gradient problem. In the experiments, MCNet reaches the overall 24-modulation classification rate of 93.59% at 20 dB SNR on the well-known DeepSig dataset.
26 April 2021 in IEEE Wireless Communications Letters
Automotive radars, with a widespread emergence in the last decade, have faced various jamming attacks. Utilizing low probability of intercept (LPI) radar waveforms, as one of the essential solutions, demands an accurate waveform recognizer at the intercept receiver. Numerous conventional approaches have been studied for LPI radar waveform recognition, but their performance is inadequate under channel condition deterioration. In this letter, by exploiting deep learning (DL) to capture intrinsic radio characteristics, we propose a convolutional neural network (CNN), namely LPI-Net, for automatic radar waveform recognition. In particular, radar signals are first analyzed by a time-frequency analysis using the Choi-Williams distribution. Subsequently, LPI-Net, primarily consisting of three sophisticated modules, is built to learn the representational features of time-frequency images, in which each module is constructed with a preceding maps collection to gain feature diversity and a skip-connection to maintain informative identity. Simulation results show that LPI-Net achieves the 13-waveform recognition accuracy of over 98% at 0 dB SNR and further performs better than other deep models.
28 November 2020 in Applied Intelligence
15 April 2019 in IEEE Transactions on Industrial Informatics
Recently, numerous methods have been introduced for 3D action recognition using handcrafted feature descriptors coupled traditional classifiers. However, they cannot learn high-level features of a whole skeleton sequence exhaustively. In this paper, a novel encoding technique, namely Pose Feature to Image (PoF2I), is introduced to transform the pose features of joint-joint distance and orientation to color pixels. By concatenating the features of all skeleton frames in a sequence, a color image is generated to depict spatial joint correlations and temporal pose dynamics of an action appearance. The strategy of end-to-end fine-tuning a pre-trained deep convolutional neural network, which completely capture multiple high-level features at multi-scale action representation, is implemented for learning recognition models. We further propose an efficient data augmentation mechanism for informative enrichment and overfitting prevention. The experimental results on six challenging 3D action recognition datasets demonstrate that the proposed method outperforms state-of-the-art approaches.
1 March 2020 in Information Sciences
2 December 2020 in IEEE Journal of Biomedical and Health Informatics
Objective: With the scenario of limited labeled dataset, this paper introduces a deep learning-based approach that leverages Diabetic Retinopathy (DR) severity recognition performance using fundus images combined with wide-field swept-source optical coherence tomography angiography (SS-OCTA). Methods: The proposed architecture comprises a backbone convolutional network associated with a Twofold Feature Augmentation mechanism, namely TFA-Net. The former includes multiple convolution blocks extracting representational features at various scales. The latter is constructed in a two-stage manner, i.e., the utilization of weight-sharing convolution kernels and the deployment of a Reverse Cross-Attention (RCA) stream. Results: The proposed model achieves a Quadratic Weighted Kappa rate of 90.2% on the small-sized internal KHUMC dataset. The robustness of the RCA stream is also evaluated by the single-modal Messidor dataset, of which the obtained mean Accuracy (94.8%) and Area Under Receiver Operating Characteristic (99.4%) outperform those of the state-of-the-arts significantly. Conclusion: Utilizing a network strongly regularized at feature space to learn the amalgamation of different modalities is of proven effectiveness. Thanks to the widespread availability of multi-modal retinal imaging for each diabetes patient nowadays, such approach can reduce the heavy reliance on large quantity of labeled visual data. Significance: Our TFA-Net is able to coordinate hybrid information of fundus photos and wide-field SS-OCTA for exhaustively exploiting DR-oriented biomarkers. Moreover, the embedded feature-wise augmentation scheme can enrich generalization ability efficiently despite learning from small-scale labeled data.