Mingzhe Zhu

Bio: Mingzhe Zhu is an academic researcher from Xidian University. The author has contributed to research in topics: Computer science & Feature (computer vision). The author has an hindex of 3, co-authored 11 publications receiving 26 citations.

Self-Matching CAM: A Novel Accurate Visual Explanation of CNNs for SAR Image Interpretation

Abstract: Synthetic aperture radar (SAR) image interpretation has long been an important but challenging task in SAR imaging processing. Generally, SAR image interpretation comprises complex procedures including filtering, feature extraction, image segmentation, and target recognition, which greatly reduce the efficiency of data processing. In an era of deep learning, numerous automatic target recognition methods have been proposed based on convolutional neural networks (CNNs) due to their strong capabilities for data abstraction and mining. In contrast to general methods, CNNs own an end-to-end structure where complex data preprocessing is not needed, thus the efficiency can be improved dramatically once a CNN is well trained. However, the recognition mechanism of a CNN is unclear, which hinders its application in many scenarios. In this paper, Self-Matching class activation mapping (CAM) is proposed to visualize what a CNN learns from SAR images to make a decision. Self-Matching CAM assigns a pixel-wise weight matrix to feature maps of different channels by matching them with the input SAR image. By using Self-Matching CAM, the detailed information of the target can be well preserved in an accurate visual explanation heatmap of a CNN for SAR image interpretation. Numerous experiments on a benchmark dataset (MSTAR) verify the validity of Self-Matching CAM.

A Novel Data-Driven Specific Emitter Identification Feature Based on Machine Cognition

Abstract: Machine learning becomes increasingly promising in specific emitter identification (SEI), particularly in feature extraction and target recognition. Traditional features, such as radio frequency (RF), pulse amplitude (PA), power spectral density (PSD), and etc., usually show limited recognition effects when only a slight difference exists in radar signals. Numerous two-dimensional features on transform domain, like various time-frequency representation and ambiguity function are used to augment information abundance, whereas the unacceptable computational burden usually emerges. To solve this problem, some artfully handcrafted features in transformed domain are proposed, like representative slice of ambiguity function (AF-RS) and compressed sensing mask (CS-MASK), to extract representative information that contributes to machine recognition task. However, most handcrafted features only utilizing neural network as a classifier, few of them focus on mining deep informative features from the perspective of machine cognition. Such feature extraction that is based on human cognition instead of machine cognition may probably miss some seemingly nominal texture information which actually contributes greatly to recognition, or collect too much redundant information. In this paper, a novel data-driven feature extraction is proposed based on machine cognition (MC-Feature) resort to saliency detection. Saliency detection exhibits positive contributions and suppresses irrelevant contributions in a transform domain with the help of a saliency map calculated from the accumulated gradients of each neuron to input data. Finally, positive and irrelevant contributions in the saliency map are merged into a new feature. Numerous experimental results demonstrate that the MC-feature can greatly strengthen the slight intra-class difference in SEI and provides a possibility of interpretation of CNN.

Compressed Sensing Mask Feature in Time-Frequency Domain for Civil Flight Radar Emitter Recognition

Abstract: Specific emitter identification (SEI) is gaining popularity since it can distinguish different individuals in same type of radar emitter under complex electromagnetic environment. However, classification of signals is still a challenging task when the feature has low physical representation. In this work, we propose a compressed sensing mask feature in ambiguity domain, which can significantly improve the recognition rate of civil flight radar emitters. Furthermore, it not only represents physical characteristics of measured radar signals but also contains more time varying information and alleviates the computational costs. The physical significance and effectiveness of the proposed feature can be verified by reconstructing Wigner-Ville distribution (WVD) from the sparsest ambiguity function. Experimental results corroborate the highly accuracy and stability of the proposed approach.

Specific Emitter Identification Based on Synchrosqueezing Transform for Civil Radar

Abstract: Time-frequency (TF) signal features are widely used in specific emitter identification (SEI) which commonly arises in many applications, especially for radar signals. Due to data scale and algorithm complexity, it is difficult to obtain an informative representation for SEI with existing TF features. In this paper, a feature extraction method is proposed based on synchrosqueezing transform (SST). The SST feature has an equivalent dimension to Fourier transform, and retains the most relevant information of the signal, leading to on average approximately 20 percent improvement in SEI for complex frequency modulation signals compared with existing handcrafted features. Numerous results demonstrate that the synchrosqueezing TF feature can offer a better recognition accuracy, especially for the signals with intricate time-varying information. Moreover, a linear relevance propagation algorithm is employed to attest to the SST feature importance from the perspective of deep learning.

An adaptive STFT using energy concentration optimization

Abstract: This work is inspired by a kind of S-transforms which employ a window width optimization strategy. Since the main objective of their methods is to adjust the window width passively by controlling the standard deviation function of the Gaussian window, we realize it in a way more straightly. Namely, we propose a new window which can optimize the shape and width directly according to a modified concentration measure. Notice that this window is not the function of frequency, so the corresponding time-frequency representation is not under the framework of S-transform. It is a kind of adaptive short-time Fourier transforms. The performance of the method is tested by a set of synthetic FM signals. The appropriate window of the proposed method can be given definitely and the method can achieve high energy concentration in low SNR.

Locally Optimized Adaptive Directional Time–Frequency Distributions

Abstract: This paper addresses the problem of estimating the parameters of adaptive directional time–frequency distributions (ADTFDs). ADTFDs locally optimize the direction of the smoothing kernel on the basis of directional Gaussian or double derivative directional Gaussian filter. Conventionally, the parameters of these techniques have to be tuned manually for each particular signal. Global optimization of the parameters fails to provide the desired results when the signal encompasses close or short-duration components. We propose a two-stage algorithm to resolve this issue. As part of the first stage, the length of the smoothing kernel is optimized globally. In the second stage, the parameters which control the shape of the selected smoothing window are optimized, locally. It is shown that the multistage algorithm can result in a time–frequency distribution that has both high resolution for close components and good concentration of signal energy for short-duration signal components. Experimental findings reveal the superiority of the proposed technique over the existing methods in the case of complete signals and its benefits in the case of signals with missing samples.

Radio Frequency Fingerprint Collaborative Intelligent Identification Using Incremental Learning

Abstract: For distributed sensor systems using neural networks, each sub-network has a different electromagnetic environment, and these recognition accuracy is also different. In this paper, we propose a distributed sensor system using incremental learning to solve the problem of radio frequency fingerprint identification. First, the intelligent representation of the received signal is linearly fused into a four-channel image. Then, convolutional neural network is trained by using the existing data to obtain the preliminary model of the network, and decision fusion is used to solve the problem in the distributed system. Finally, using new data, instead of retraining the model, we employ incremental learning by fine-tuning the preliminary model. The proposed method can significantly reduce the training time and is adaptive to streaming data. Extensive experiments show that the proposed method is computationally efficient, and also has satisfactory recognition accuracy, especially at low signal-to-noise ratio (SNR) regime.

Self-Matching CAM: A Novel Accurate Visual Explanation of CNNs for SAR Image Interpretation

Abstract: Synthetic aperture radar (SAR) image interpretation has long been an important but challenging task in SAR imaging processing. Generally, SAR image interpretation comprises complex procedures including filtering, feature extraction, image segmentation, and target recognition, which greatly reduce the efficiency of data processing. In an era of deep learning, numerous automatic target recognition methods have been proposed based on convolutional neural networks (CNNs) due to their strong capabilities for data abstraction and mining. In contrast to general methods, CNNs own an end-to-end structure where complex data preprocessing is not needed, thus the efficiency can be improved dramatically once a CNN is well trained. However, the recognition mechanism of a CNN is unclear, which hinders its application in many scenarios. In this paper, Self-Matching class activation mapping (CAM) is proposed to visualize what a CNN learns from SAR images to make a decision. Self-Matching CAM assigns a pixel-wise weight matrix to feature maps of different channels by matching them with the input SAR image. By using Self-Matching CAM, the detailed information of the target can be well preserved in an accurate visual explanation heatmap of a CNN for SAR image interpretation. Numerous experiments on a benchmark dataset (MSTAR) verify the validity of Self-Matching CAM.

A Novel Data-Driven Specific Emitter Identification Feature Based on Machine Cognition

Abstract: Machine learning becomes increasingly promising in specific emitter identification (SEI), particularly in feature extraction and target recognition. Traditional features, such as radio frequency (RF), pulse amplitude (PA), power spectral density (PSD), and etc., usually show limited recognition effects when only a slight difference exists in radar signals. Numerous two-dimensional features on transform domain, like various time-frequency representation and ambiguity function are used to augment information abundance, whereas the unacceptable computational burden usually emerges. To solve this problem, some artfully handcrafted features in transformed domain are proposed, like representative slice of ambiguity function (AF-RS) and compressed sensing mask (CS-MASK), to extract representative information that contributes to machine recognition task. However, most handcrafted features only utilizing neural network as a classifier, few of them focus on mining deep informative features from the perspective of machine cognition. Such feature extraction that is based on human cognition instead of machine cognition may probably miss some seemingly nominal texture information which actually contributes greatly to recognition, or collect too much redundant information. In this paper, a novel data-driven feature extraction is proposed based on machine cognition (MC-Feature) resort to saliency detection. Saliency detection exhibits positive contributions and suppresses irrelevant contributions in a transform domain with the help of a saliency map calculated from the accumulated gradients of each neuron to input data. Finally, positive and irrelevant contributions in the saliency map are merged into a new feature. Numerous experimental results demonstrate that the MC-feature can greatly strengthen the slight intra-class difference in SEI and provides a possibility of interpretation of CNN.

Specific Emitter Identification Based on Synchrosqueezing Transform for Civil Radar

Abstract: Time-frequency (TF) signal features are widely used in specific emitter identification (SEI) which commonly arises in many applications, especially for radar signals. Due to data scale and algorithm complexity, it is difficult to obtain an informative representation for SEI with existing TF features. In this paper, a feature extraction method is proposed based on synchrosqueezing transform (SST). The SST feature has an equivalent dimension to Fourier transform, and retains the most relevant information of the signal, leading to on average approximately 20 percent improvement in SEI for complex frequency modulation signals compared with existing handcrafted features. Numerous results demonstrate that the synchrosqueezing TF feature can offer a better recognition accuracy, especially for the signals with intricate time-varying information. Moreover, a linear relevance propagation algorithm is employed to attest to the SST feature importance from the perspective of deep learning.