A new end-to-end image dehazing algorithm based on residual attention mechanism

Traditional image dehazing algorithms based on prior knowledge and deep learning rely on the atmospheric scattering model and are easy to cause color distortion and incomplete dehazing. To solve these problems, an end-to-end image dehazing algorithm based on residual attention mechanism is proposed in this paper. The network includes four modules: encoder, multi-scale feature extraction, feature fusion and decoder. The encoder module encodes the input haze image into feature map, which is convenient for subsequent feature extraction and reduces memory consumption; the multi-scale feature extraction module includes residual smoothed dilated convolution module, residual block and efficient channel attention, which can expand the receptive field and extract different scale features by filtering and weighting; the feature fusion module with efficient channel attention adjusts the channel weight dynamically, acquires rich context information and suppresses redundant information so as to enhance the ability to extract haze density image of the network; finally, the encoder module maps the fused feature nonlinearly to obtain the haze density image and then restores the haze free image. The qualitative and quantitative tests based on SOTS test set and natural haze images show good objective and subjective evaluation results. This algorithm improves the problems of color distortion and incomplete dehazing effectively.

Download Full-text

Deep learning-based tool wear prediction and its application for machining process using multi-scale feature fusion and channel attention mechanism

Measurement ◽

10.1016/j.measurement.2021.109254 ◽

2021 ◽

Vol 177 ◽

pp. 109254

Author(s):

Xingwei Xu ◽

Jianwen Wang ◽

Bingfu Zhong ◽

Weiwei Ming ◽

Ming Chen

Keyword(s):

Deep Learning ◽

Tool Wear ◽

Feature Fusion ◽

Attention Mechanism ◽

Machining Process ◽

Wear Prediction ◽

Scale Feature ◽

Multi Scale ◽

Tool Wear Prediction

Download Full-text

Image Compressive Sensing via Multi-scale Feature Extraction and Attention Mechanism

2020 International Conference on Intelligent Computing, Automation and Systems (ICICAS) ◽

10.1109/icicas51530.2020.00061 ◽

2020 ◽

Author(s):

Chuning He

Keyword(s):

Feature Extraction ◽

Compressive Sensing ◽

Attention Mechanism ◽

Scale Feature ◽

Multi Scale

Download Full-text

FD-GAN: Generative Adversarial Networks with Fusion-Discriminator for Single Image Dehazing

Proceedings of the AAAI Conference on Artificial Intelligence ◽

10.1609/aaai.v34i07.6701 ◽

2020 ◽

Vol 34 (07) ◽

pp. 10729-10736 ◽

Cited By ~ 4

Author(s):

Yu Dong ◽

Yihao Liu ◽

He Zhang ◽

Shifeng Chen ◽

Yu Qiao

Keyword(s):

Training Data ◽

Training Dataset ◽

Generative Adversarial Networks ◽

Single Image ◽

Image Dehazing ◽

Atmospheric Scattering ◽

Color Distortion ◽

Adversarial Networks ◽

End To End ◽

Single Image Dehazing

Recently, convolutional neural networks (CNNs) have achieved great improvements in single image dehazing and attained much attention in research. Most existing learning-based dehazing methods are not fully end-to-end, which still follow the traditional dehazing procedure: first estimate the medium transmission and the atmospheric light, then recover the haze-free image based on the atmospheric scattering model. However, in practice, due to lack of priors and constraints, it is hard to precisely estimate these intermediate parameters. Inaccurate estimation further degrades the performance of dehazing, resulting in artifacts, color distortion and insufficient haze removal. To address this, we propose a fully end-to-end Generative Adversarial Networks with Fusion-discriminator (FD-GAN) for image dehazing. With the proposed Fusion-discriminator which takes frequency information as additional priors, our model can generator more natural and realistic dehazed images with less color distortion and fewer artifacts. Moreover, we synthesize a large-scale training dataset including various indoor and outdoor hazy images to boost the performance and we reveal that for learning-based dehazing methods, the performance is strictly influenced by the training data. Experiments have shown that our method reaches state-of-the-art performance on both public synthetic datasets and real-world images with more visually pleasing dehazed results.

Download Full-text

A Novel 2D-3D CNN with Spectral-Spatial Multi-Scale Feature Fusion for Hyperspectral Image Classification

Remote Sensing ◽

10.3390/rs13224621 ◽

2021 ◽

Vol 13 (22) ◽

pp. 4621

Author(s):

Dongxu Liu ◽

Guangliang Han ◽

Peixun Liu ◽

Hang Yang ◽

Xinglong Sun ◽

...

Keyword(s):

Feature Extraction ◽

Image Classification ◽

Classification Scheme ◽

Hyperspectral Image ◽

Feature Fusion ◽

Spectral Feature ◽

Hyperspectral Image Classification ◽

Scale Feature ◽

Multi Scale ◽

3D Cnn

Multifarious hyperspectral image (HSI) classification methods based on convolutional neural networks (CNN) have been gradually proposed and achieve a promising classification performance. However, hyperspectral image classification still suffers from various challenges, including abundant redundant information, insufficient spectral-spatial representation, irregular class distribution, and so forth. To address these issues, we propose a novel 2D-3D CNN with spectral-spatial multi-scale feature fusion for hyperspectral image classification, which consists of two feature extraction streams, a feature fusion module as well as a classification scheme. First, we employ two diverse backbone modules for feature representation, that is, the spectral feature and the spatial feature extraction streams. The former utilizes a hierarchical feature extraction module to capture multi-scale spectral features, while the latter extracts multi-stage spatial features by introducing a multi-level fusion structure. With these network units, the category attribute information of HSI can be fully excavated. Then, to output more complete and robust information for classification, a multi-scale spectral-spatial-semantic feature fusion module is presented based on a Decomposition-Reconstruction structure. Last of all, we innovate a classification scheme to lift the classification accuracy. Experimental results on three public datasets demonstrate that the proposed method outperforms the state-of-the-art methods.

Download Full-text