Thick cloud cover can completely obscure the surface of the earth from satellite view, while thinner haze and shadows distort the image of rural and urban regions. As such, many remote sensing images for monitoring climate, crops, and urban growth are only partially usable.
Research in the International Journal of Bio-Inspired Computation offers a way for satellites to see through clouds using a hybrid artificial intelligence system. The system essentially removes clouds from the images sent back by the satellite and reconstructs the land surface beneath with greater fidelity than is possible with earlier techniques. Almost all optical satellite images are affected by clouds to some degree, so improvements in AI cloud removal could expand the reliability of high-resolution Earth observation data.
Traditional approaches have relied either on physical models of atmospheric light scattering or on image-processing techniques that compare multiple images through time or across different wavelengths of light. Those methods are useful but struggle with varying cloud thickness or large, fully obscured areas. More recent machine learning systems, in which algorithms learn patterns from large datasets, have improved results, but they need clear reference images, without them, they simply produce blurred areas where the landscape was obscured by clouds.
The new approach is a deep denoising application known as SenseNet. It treats those image pixels with clouds or haze as being structured noise that can be removed. The system uses a model inspired by nature called a hybrid Coyote Fox Optimisation algorithm, which works by modelling the social, cooperative behaviour in canines to take the input data and process it to find the optimal solution. In computational terms, it helps tune the network’s internal parameters so that training does not stall on suboptimal solutions that would otherwise confound the learning algorithm.
Compared with existing denoising approaches, the system improved signal-to-noise ratios by more than two decibels and reduced residual errors. An improvement of just 2 dB is an almost 60 per cent improvement.
By clearing the clouds away, the system can more readily delineate agricultural boundaries and map road networks and bodies of water so that phenomena such as deforestation, crop yields, and infrastructure can be viewed with more detail. In persistently cloudy regions, including much of the tropics, more reliable cloud removal could reduce data gaps, supporting climate adaptation and disaster response strategies that increasingly depend on near-real-time satellite intelligence.
Gound, R.S. and Thepade, S.D. (2026) ‘SenseNet: satellite image enhancement using optimised deep denoiser for cloud removal’, Int. J. Bio-Inspired Computation, Vol. 27, No. 1, pp.45–59.
No comments:
Post a Comment