TRENDING RESEARCH

Deep Learning Approaches for Mine Water Inflow Management 

Mining ores in search of valuable geological minerals like, coal, gold, iron etc. play a vital role in improving the country’s economy. However, these ores normally occur below ground water levels that a sudden or managed water inflow from surrounding aquifers or any water-containing rock fissures or goafs from where minerals have been already excavated, tend to disrupt mining. The result is the heavy mining production loss, the question on safety and the poor water management in and around the mining panel. Hence, controlled water inflows from the water inlets and its proper draining from the mining regions is much preferred to increase the mining productivity as well as safety, in addition to ensuring stable ground water levels and ecological sustainability. There are methods that mathematically formulate controlled water inflow and drainage to aid productive mining, especially based on the knowledge from mining plans as well as design and various physical processes employed in it. Yet, when the mining depths grow larger and the mining panel’s shape become unpredictably irregular, these mathematical formulations and the associated geological parameters become more complex and region-centric for application!!! This is not the stop…The deep learning models can certainly render support at this juncture because of their increased training efficiency and accurate time-variable forecast capability, not wanting to consider all the geological parameters affecting water inflow. Researchers from China and Vietnam have predicted an accurate water inflow for the subsequent day to prevent water damage from an aquifer in a coal roof in Tingnan Coal Mine, located in Changwu county, central and western Shanxi Province, China. They have achieved this by applying deep learning approaches on the groundwater level data from hydrological borehole monitoring, advance mining distance and daily water inflow data. Despite saying “The developed deep learning models could support for water treatment and environment management in mining areas”, the researchers desire the future research to centre around longer time-ahead forecast, rather than on a one-day forecast. The researchers point that considering more factors like, micro seismic monitoring data or stress strain and fracture expansion measurements could improve the model’s hysteresis and predictive accuracy, paving way for upcoming researchers to contribute in water inflow management, while monitoring and alerting about water inrush disasters too…

Image credit: www.freepik.com

Underwater Image Enhancement for Better Object Tracking and Detection

 Underwater exploration will be effective, if the images of different objects and the various colourful flora or fauna under water looks crystal clear in a more splendid and colourful format, as does the original being captured by a camera. Unfortunately, water and its associated constituents highly attenuates, absorbs and scatters the different wavelengths of light that can bring the real picture of its objects to light!!! The outcome is hence a colour-changed, hazed, noisy, blurred and contrast as well as brightness-limited image of the object under water, making it difficult for object tracking and detection under water in varied fields of application like, remote sensing, military, civil, ocean research and many more. Hence, the research on enhancing the underwater image either in the form of image quality enhancement, image restoration, image deblurring, image contrast enhancement, image dehazing, image colour conversion and image denoising has become a necessity and varied approaches have been put forth incessantly to aid in varied applications. The combined research from China and Australia, being published in Engineering Applications of Artificial Intelligence, Elsevier, Vol: 121, is one such kind that showcases a multi-feature underwater image enhancement method via embedded fusion mechanism (MFEF). Utilizing the UIEB dataset, the corresponding researchers have rendered a proper white- balanced and contrast-equalized input image for yielding better underwater image reconstruction with improved sharpness, colour saturation  and image detail. “Our method tests the applicability of the method with saliency detection and feature matching, which provides practical implications for advanced computer vision tasks”, the researchers say. Though their research has shown potential improvements, a cent percent reproduction of the reality from underwater images will be the driving force for future research.

Image credit: www. unsplash.com

Cardiac Imaging-based Prediction Models to Improve Diagnosis

It is very common now to hear that a person, who we met an hour or minute before, is no more at present and sadly the cause is the cardiovascular ailments. Numerous clinical advancements to diagnose these ailments at its initial phases to save lives have been developed in the past few decades and one among them is the cardiac imaging-based model generation using machine learning approaches and expert assistance. With these image-based models, the heart and its associated blood vessels that pump blood cyclically to all bodily parts can be non-invasively checked for their structure or function in an in vivo manner. The outcome is the easy and rapid diagnosis of a particular cardio vascular disease, rendering complete assistance in performing surgery or planning a medical treatment. The use of machine learning approaches has indeed resulted in accurate diagnosis in a patient-unspecific sense, enabling large number of patients to be handled in less time. However, as with all imaging-based applications, here too the cardiac images are subjected to imaging artifacts and low image contrast or resolution, affecting accurate diagnosis. Further, the heart and its associated blood vessels are very complexly -interlinked that their tissue discontinuities or jaggedness can falsify the image-based prediction results, which is highly unagreeable. Hence, research to produce or improve cardiac imaging-based predictive models is always encouraged. One of the USA-based researches supports this notion by taking efforts to generate the left ventricle geometry from short-axis MR images using deep learning approaches, being guided anatomically. Proving that their model could generate smooth, continuous, and accurate anatomical features of left ventricle, the researchers convey about the adaptability and applicability of their models to other organs like, liver, kidney or other big blood vessels in the near future. In all, it is in the hands of researchers and medical personnel to go hand in hand for developing approaches, which can cut down the quickly-increasing morbidity and mortality rates.

Image Credit: www.freepik.com