Researchers in China have developed an AI technique to improve the quality of underwater image sensors, writes Nick Flaherty. The team at the Hefei Institutes of Physical Science of the Chinese Academy of Sciences created the Learnable Fullfrequency Transformer Dual Generative Adversarial Network (LFT-DGAN) to address the issue of underwater image-quality degradation. This can be caused by variations in colour, low contrast and signal noise from the changing density of the water. Enhancement technology aims to optimise the quality of underwater images and meet the diverse needs of marine scientific research, underwater robots and object recognition. A reversible convolution image decomposition technique separates underwater picture information into low-, medium-, and high-frequency domains, enabling more thorough feature extraction. These are separated out into image channels, and the spatial similarity of the pictures is used to construct a learnable, full-frequency, domain-transformer network. This transformer facilitates interaction between different branches of information, enhancing overall image-processing capabilities. A dual-domain discriminator is capable of learning the spatial and frequency domain characteristics of underwater images. Here, the adversarial neural networks pit two different types of AI network against each other to produce the best output. The effectiveness of the resulting dual-generative adversarial neural-network model of the full-frequency transformer was verified by comparing multiple underwater image experimental data. With the help of this model, researchers used image decomposition technology with reversible convolution for the first time to accurately separate the different frequency features of the image. Artificial intelligence Enhancing underwater photography with AI Dual adversarial neural networks improve the quality of underwater images (Image courtesy of Hefei Institutes of Physical Science) www.tronics.tdk.com Tactical grade. Low SWaP. Digital. © Adobe Stock MEMS inertial sensors for land, sea, and air.
RkJQdWJsaXNoZXIy MjI2Mzk4