Xin Chen

Addressing the limitations of conventional gas leak detection techniques in petrochemical environments, this study introduces a deep learning framework (DSEF-CNN) based on Fourier transform infrared spectroscopy. Traditional approaches such as multilayer perceptron (MLP) and 1D-CNN exhibit reduced recognition accuracy. This is due to spectral overlap among multiple gases, industrial noise interference, and their low sensitivity to low-concentration leaks. The DSEF-CNN is designed specifically for high-resolution FTIR spectral data and integrates four targeted components: (1) a differential processing module amplifies dynamic concentration changes during leaks, which leverages FTIR’s subtle spectral capture to boost sensitivity to low-concentration leaks; (2) a 1D feature pyramid network fuses multi-scale features, from fine fingerprint peaks to broad functional group peaks, which resolves spectral overlap and aligns with FTIR’s wavenumber-dependent characteristics; (3) an efficient channel attention mechanism selectively emphasizes wavenumber channels related to target gases, which mitigates industrial noise in complex FTIR profiles; and (4) a lightweight CNN backbone enables end-to-end classification, which is optimized for real-time processing of continuous FTIR spectral streams. Extensive simulations and field experiments demonstrate that the DSEF-CNN model significantly improves recognition accuracy. It is thereby validated for real-time multi-gas leak identification in petrochemical monitoring scenarios.

This article proposes a novel hyperspectral image (HSI) denoising and destriping method based on graph signal processing (GSP) that fully exploits the HSI properties. Double graph Laplacian regularizers (DGLR) are established to illustrate the spectral and spatial connections of an HSI. An HSI is of low rank property, and its spectral differences exhibit a lower rank than the HSI itself. To characterize this property, we construct our first graph by using the difference between every two adjacent bands as the graph nodes. In addition, to preserve the spatial connection of the HSIs and avoid over-smoothing, the second graph takes the full-band HSI blocks as graph nodes. We use the Euclidean distance between nodes and the $\ell _{2}$ norm between graph signals to compute the edge weights of these two graphs. Besides, we also consider removing the sparse/impulse noise and stripe noise by using an $\ell _{1}$ norm and a nuclear norm as constraints. An augmented Lagrange multiplier method is developed to solve the proposed optimization problem. To testify the competitive performance of our method, we conduct numerous experiments on both simulated noisy HSIs and real noisy HSIs. The quantitative evaluation indices and visual performance show that DGLR surpasses many mainstream denoising methods within less or comparable time.

In recent years, target tracking has made great progress in accuracy. This development is mainly attributed to powerful networks (such as transformers) and additional modules (such as online update and refinement modules). However, less attention has been paid to tracking speed. Most state-of-the-art trackers are satisfied with the real-time speed on powerful GPUs. However, practical applications necessitate higher requirements for tracking speed, especially when edge platforms with limited resources are used. In this work, we present an efficient tracking method via a hierarchical cross-attention transformer named HCAT. Our model runs about 195 fps on GPU, 45 fps on CPU, and 55 fps on the edge AI platform of NVidia Jetson AGX Xavier. Experiments show that our HCAT achieves promising results on LaSOT, GOT-10k, TrackingNet, NFS, OTB100, UAV123, and VOT2020. Code and models are available at https://github.com/chenxin-dlut/HCAT.

Hyperspectral image (HSI) is easily corrupted by different kinds of noise, such as stripes, dead pixels, impulse noise, Gaussian noise, etc. Due to less consideration of the structural specificity of stripes, many existing HSI denoising methods cannot effectively remove the heavy stripes in mixed noise. In this paper, we classify the noise on HSI into three types: sparse noise, stripe noise, and Gaussian noise. The clean image and different types of noise are treated as independent components. In this way, the image denoising task can be naturally regarded as an image decomposition problem. Thanks to the structural characteristic of stripes and the low-rank property of HSI, we propose to destripe and denoise the HSI by using stripe and spectral low-rank matrix recovery and combine it with the global spatial-spectral TV regularization (SSLR-SSTV). By considering different properties of different HSI ingredients, the proposed method separates the original image from the noise components perfectly. Both simulation and real image denoising experiments demonstrate that the proposed method can achieve a satisfactory denoising result compared with the state-of-the-art methods. Especially, it outperforms the other methods in the task of stripe noise removal visually and quantitatively.