Research on facial expression recognition (FER) technology can promote the development of theoretical and practical applications for our daily life. Currently, most of the related works on this technology are focused on un-occluded FER. However, in real life, facial expression images often have partial occlusion; therefore, the accurate recognition of occluded facial expression images is a topic that should be explored. In this paper, we proposed a novel Wasserstein generative adversarial network-based method to perform occluded FER. After complementing the face occlusion image with complex facial expression information, the recognition is achieved by learning the facial expression features of the images. This method consists of a generator G and two discriminators D ₁ and D ₂ . The generator naturally complements occlusion in the expression image under the triple constraints of weighted reconstruction loss l _wr , triplet loss l _t , and adversarial loss l _a . We optimize the discriminator D ₁ to distinguish between real and fake by constructing an adversarial loss l _a between the generated complementing images, original un-occluded images, and small-scale-occluded images based on the Wasserstein distance. Finally, the FER is completed by introducing classification loss l _c into D ₂ . To verify the effectiveness of the proposed method, an experimental analysis was performed on the AffectNet and RAF-DB datasets. The visual occlusion complementing results, comparison of recognition rates of facial expression images with and without de-occlusion processing, and T-distributed stochastic neighbor embedding visual analysis of facial expression features all prove the effectiveness of the proposed method. The experimental results show that the proposed method is better than the existing state-of-the-art methods.

Fatigued driving detection in complex environments is a challenging problem. This paper proposes a fatigued driving detection algorithm based on multi-index fusion and a state recognition network, for further analysis of driver fatigue states. This study uses a multi-task cascade convolutional neural network for face detection and facial key point detection, corrects the face according to the key points of the eye, intercepts a binoculus image to recognize the eye state, and intercepts a mouth image according to the left and right corner points to recognize the mouth state. This can improve the detection accuracy of the driver's head tilt, deflection, and so on. Next, an eye state recognition network is constructed for the binoculus image to identify the eye closure state, and a mouth state recognition network is used to identify the mouth state. Finally, a fatigue judgment model is established by combining the two characteristics of the eye state and the mouth state to further analyze the driver fatigue state. The algorithm achieved 98.42% detection accuracy on a public eye dataset and achieved 97.93% detection accuracy on an open mouth dataset. As compared with other existing algorithms, the proposed algorithm has the advantages of high accuracy and simple implementation.

Change detection has recently become a topic of great significance in the field of remote sensing. However, as one of the traditional effective detection methods, synthetic aperture radar (SAR) image change detection based on wavelet transform fusion remains limited because of the existence of speckle noise and because multidirectional information has been underutilized. Therefore, we propose an unsupervised change detection method using saliency extraction and the shearlet transform. Saliency extraction is first used to homogenize registered images to reduce speckle noise. Using a subtraction operation for preprocessed binary images, we then obtain a saliency-guided difference image (DI) that includes the main contour change information. Then, the Gauss-log-ratio DI includes the detailed change information at the edge of the image and acts as an auxiliary DI. Next, two DIs are fused with the shearlet transform. During this process, the DI is decomposed into one low-frequency and four high-frequency subimages. The low-frequency subimage contains image contour information and the high-frequency subimages contain the edge information of the images. Compared to wavelet fusion, in our method, no extra fusion noise occurs because the shearlet transform performs multiscale analysis. The final change map can be obtained through maximum entropy segmentation. Real SAR image pairs in areas of Bern, Switzerland, and Suzhou, China, are used to verify the proposed change detection method. The experimental results demonstrate the effectiveness of the proposed method when compared to the reference methods.

In this paper, we propose to use deep neural networks for image compression in the wavelet transform domain. When the input image is transformed from spatial pixel domain to the wavelet transform domain, one low frequency sub-band (LF sub-band) and three high frequency sub-bands (HF sub-bands) are generated. Low frequency sub-band is firstly used to predict each high frequency sub-band to eliminate redundancy between sub-bands, after which the sub-bands are fed into different auto-encoders to do the encoding. In order to further improve the compression efficiency, we use a conditional probability model to estimate the context-dependent prior probability of the encoded codes, which can be used for entropy coding. The entire training process is unsupervised, and the auto-encoders and the conditional probability model are trained jointly. Experimental results show that the proposed approach outperforms JPEG, JPEG2000, BPG and some mainstream neural network based image compression. Furthermore, it produces a better visual quality with clearer details and textures because more high frequency coefficients can be reserved thanks to the high frequency prediction.

Suppressing various noises while achieving precise signal reconstruction in Distributed Acoustic Sensing Vertical Seismic Profiling (DAS VSP) remains a challenge. Existing denoising methods are insufficient due to factors such as the unknown noise-disturbing mechanism, low SNR, and limited training data. Therefore, this study proposes the Mean-Stochastic Differential Equation (SDE) diffusion model as an advanced solution. Built upon the standard diffusion model, which incorporates forward and backward diffusion processes, our model introduced three modifications to enhance performance. 1. Improving the forward diffusion process: Transforming the final state into a combination of the noisy DAS VSP and Gaussian noise. This adjustment allows precise representations of multi-type noise generation and facilitates backward sampling. 2. Enhancing noise prediction between successive steps in the backward process: A Nonlinear Activation Free Network (NAFnet) with a time Multi-Layer Perceptron (MLP) was employed to provide accurate noise predictions at different states. 3. Addressing training instability inherent in standard diffusion: The objective function is modified to seek the optimal trajectory of the best quality of signal reconstruction rather than directly evaluating the noise prediction. The forward diffusion is a dynamic evolution of adding noise to the pure signal, while the backward processing aims to remove the noise step by step. Comprehensive experiments demonstrate the superiority of our method in diverse noise suppression, signal resolution enhancement, and amplitude preservation. Moreover, grounded in physics-based equations, our method exhibits less dependency on training data compared to conventional deep learning methods.