Search Results (5)

Face recognition (FR) in unconstrained conditions remains an open research topic and an ongoing challenge. The facial images exhibit diverse expressions, occlusions, variations in illumination, and heterogeneous backgrounds. This work aims to produce an accurate and robust system for enhanced Security and Surveillance. A parallel hybrid deep learning model for feature extraction and classification is proposed. An ensemble of three parallel extraction layer models learns the best representative features using CNN and RNN. 2D LBP and 3D Mesh LBP are computed on face images to extract image features as input to two RNNs. A stacked autoencoder (SAE) merged the feature vectors extracted from the three CNN-RNN parallel layers. We tested the designed 2D/3D CNN-RNN framework on four standard datasets. We achieved an accuracy of $98.9\%$. The hybrid deep learning model significantly improves FR against similar state-of-the-art methods. The proposed model was also tested on an unconstrained conditions human crowd dataset, and the results were very promising with an accuracy of $95\%$. Furthermore, our model shows an 11.5% improvement over similar hybrid CNN-RNN architectures, proving its robustness in complex environments where the face can undergo different transformations. Full article

The performance of near-field acoustic holography (NAH) with a sparse sampling rate will be affected by spatial aliasing or inverse ill-posed equations. Through a 3D convolution neural network (CNN) and stacked autoencoder framework (CSA), the data-driven CSA-NAH method can solve this problem by utilizing the information from data in each dimension. In this paper, the cylindrical translation window (CTW) is introduced to truncate and roll out the cylindrical image to compensate for the loss of circumferential features at the truncation edge. Combined with the CSA-NAH method, a cylindrical NAH method based on stacked 3D-CNN layers (CS3C) for sparse sampling is proposed, and its feasibility is verified numerically. In addition, the planar NAH method based on the Paulis–Gerchberg extrapolation interpolation algorithm (PGa) is introduced into the cylindrical coordinate system, and compared with the proposed method. The results show that, under the same conditions, the reconstruction error rate of the CS3C-NAH method is reduced by nearly 50%, and the effect is significant. Full article

Crop type identification is the initial stage and an important part of the agricultural monitoring system. It is well known that synthetic aperture radar (SAR) Sentinel-1A imagery provides a reliable data source for crop type identification. However, a single-temporal SAR image does not contain enough features, and the unique physical characteristics of radar images are relatively lacking, which limits its potential in crop mapping. In addition, current methods may not be applicable for time-series SAR data. To address the above issues, a new crop type identification method was proposed. Specifically, a farmland mask was firstly generated by the object Markov random field (OMRF) model to remove the interference of non-farmland factors. Then, the features of the standard backscatter coefficient, Sigma-naught (σ₀), and the normalized backscatter coefficient by the incident angle, Gamma-naught (γ₀), were extracted for each type of crop, and the optimal feature combination was found from time-series SAR images by means of Jeffries-Matusita (J-M) distance analysis. Finally, to make efficient utilization of optimal multi-temporal feature combination, a new network, the convolutional-autoencoder neural network (C-AENN), was developed for the crop type identification task. In order to prove the effectiveness of the method, several classical machine learning methods such as support vector machine (SVM), random forest (RF), etc., and deep learning methods such as one dimensional convolutional neural network (1D-CNN) and stacked auto-encoder (SAE), etc., were used for comparison. In terms of quantitative assessment, the proposed method achieved the highest accuracy, with a macro-F1 score of 0.9825, an overall accuracy (OA) score of 0.9794, and a Kappa coefficient (Kappa) score of 0.9705. In terms of qualitative assessment, four typical regions were chosen for intuitive comparison with the sample maps, and the identification result covering the study area was compared with a contemporaneous optical image, which indicated the high accuracy of the proposed method. In short, this study enables the effective identification of crop types, which demonstrates the importance of multi-temporal radar images in feature combination and the necessity of deep learning networks to extract complex features. Full article

A correct food tray sealing is required to preserve food properties and safety for consumers. Traditional food packaging inspections are made by human operators to detect seal defects. Recent advances in the field of food inspection have been related to the use of hyperspectral imaging technology and automated vision-based inspection systems. A deep learning-based approach for food tray sealing fault detection using hyperspectral images is described. Several pixel-based image fusion methods are proposed to obtain 2D images from the 3D hyperspectral image datacube, which feeds the deep learning (DL) algorithms. Instead of considering all spectral bands in region of interest around a contaminated or faulty seal area, only relevant bands are selected using data fusion. These techniques greatly improve the computation time while maintaining a high classification ratio, showing that the fused image contains enough information for checking a food tray sealing state (faulty or normal), avoiding feeding a large image datacube to the DL algorithms. Additionally, the proposed DL algorithms do not require any prior handcraft approach, i.e., no manual tuning of the parameters in the algorithms are required since the training process adjusts the algorithm. The experimental results, validated using an industrial dataset for food trays, along with different deep learning methods, demonstrate the effectiveness of the proposed approach. In the studied dataset, an accuracy of 88.7%, 88.3%, 89.3%, and 90.1% was achieved for Deep Belief Network (DBN), Extreme Learning Machine (ELM), Stacked Auto Encoder (SAE), and Convolutional Neural Network (CNN), respectively. Full article

Automatic modulation classification (AMC) is playing an increasingly important role in spectrum monitoring and cognitive radio. As communication and electronic technologies develop, the electromagnetic environment becomes increasingly complex. The high background noise level and large dynamic input have become the key problems for AMC. This paper proposes a feature fusion scheme based on deep learning, which attempts to fuse features from different domains of the input signal to obtain a more stable and efficient representation of the signal modulation types. We consider the complementarity among features that can be used to suppress the influence of the background noise interference and large dynamic range of the received (intercepted) signals. Specifically, the time-series signals are transformed into the frequency domain by Fast Fourier transform (FFT) and Welch power spectrum analysis, followed by the convolutional neural network (CNN) and stacked auto-encoder (SAE), respectively, for detailed and stable frequency-domain feature representations. Considering the complementary information in the time domain, the instantaneous amplitude (phase) statistics and higher-order cumulants (HOC) are extracted as the statistical features for fusion. Based on the fused features, a probabilistic neural network (PNN) is designed for automatic modulation classification. The simulation results demonstrate the superior performance of the proposed method. It is worth noting that the classification accuracy can reach 99.8% in the case when signal-to-noise ratio (SNR) is 0 dB. Full article