Embedded Files
(Volume: 2, Issue: 3)
Speech Emotion Recognition to govern HCIs in future
The modern-day devices are of little worth, if they do not entail entities supporting Human-Computer Interaction (HCI). All the users desire to communicate with their devices with very much ease, as they communicate with other humans. Speech can be an important modality to serve this purpose, as it carries varied kinds of emotions like, happiness, sadness, angriness and many other combined emotions in it. Recognizing emotions from speech can have a massive impact on the current era of digital signal processing and HCIs. Will it not be pleasant, if a music system plays music according to your emotion in speech? Will it not be a great help to an individual, who is warned that his/ her bed-ridden loved one needs a timely support through speech emotion recognition? Will it not be a great aid to the providers, if they are informed about the customer’s mindset in getting a product/ service through the emotions carried in their speech? Will it not be an assistance to the educationalists to know the understanding ability of their students from their speech? Several researchers have given their contribution on Speech Emotion Recognition (SER) for one or more of these reasons. The key issue they faced was the optimal selection of features, might it be local or global and time, frequency or spectral-based. Md. Rayhan Ahmed, Salekul Islam, A. K. M. Muzahidul Islam and Swakkhar Shatabda from United International University, Bangladesh have shown interest on the extraction of local and long-term global contextual representations of speech signals. Utilizing five publicly available benchmark datasets (TESS, EMO-DB, RAVDESS, SAVEE and CREMA-D), involving speech in German and English, the researchers have introduced an ensemble approach that resulted from 1- D Convolutional Neural Networks, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). Despite achieving the weighted average accuracy of the state- of- the- art models, the researchers opt for real-time emotion recognition from speech in future. Hence, emotion recognition from speech signal has a great scope and renders a wide platform for researchers in the field of medicine and healthcare, marketing, education and all sorts of HCIs in the upcoming years.
Image Courtesy: www.freepik.com
AI-Based Deep Learning Approaches for Wheat Rust Monitoring
Wheat, the most consumed staple food after rice and maize, is facing an immense destruction in the crop yield due to a fungi-generated disease. This disease, called as the wheat rust disease, is becoming more virulent to affect larger number of crops because of heavy winds nowadays. Knowing the severity of the disease, do you know that the Food and Agriculture Organization of the United Nations (FAO), the Ministry of Agriculture and Forestry of the Republic of Türkiye, the International Centre for Agricultural Research in Dry Areas (ICARDA), the International Maize and Wheat Improvement Centre (CIMMYT) and the Aegean Agricultural Research Institute Directorate have conducted a 10-day workshop on May 2023 to assist the researchers in monitoring and managing the disease? So, regular monitoring of the crops become an umpteen need to ensure that ample food is served in a region, without causing crop loss due to wheat rust. But, is it not a arduous process to go behind each wheat crop for identifying the disease, where a single locality might have infinite number of crops? Many thanks to the Artificial Intelligence- based Computer Vision approaches, which could greatly help in this regard. Three researchers from Indian Council of Agricultural Research (ICAR) have collected about 6556 images of Wheat crops infected with rust in a span of three years. Naming the dataset as “WheatRust21”, they have classified the images to be healthy, brown- rust, yellow- rust and stem-rust using classical Convolutional Neural Networks- based models and eight variants of EfficientNet. Yielding a testing accuracy of 99.35%, the researchers say: “This model can be easily integrated into mobile applications for use by stakeholders for image-based wheat disease identification in field conditions”. However, they also claim that future research should incorporate a greater number of classes based on wheat rust severity.
Image courtesy: www.fao.org
CNNs to Diagnose Rotator Cuff Tears
Have you witnessed people saying, “I had a slight pain in the shoulder for few years, but now it is impossible for me to move the hands”? If so, know it that the rotator cuff tear is one of the few important causes that instigate shoulder impairments. Rotator cuff points to a group of muscles and connective tissues (tendons), which are responsible for holding the upper humerus (arm bone) ball in the shoulder blade (scapula) socket, enabling smooth movement of the arms. The cause of the tear in rotator cuff can be any of the following: (1) decreased blood flow; (2) over usage of arms and heavy weight-lifting and (3) bony spurs or bony growths, disrupting the rotator cuff muscles and causing tears in it. So, it is obvious that the athletes, the age-old people, carpenters, mechanics and other people, who give little rest to their arms during their work period are much prone to this muscle tear. As per “Rotator Cuff Injury” by Todd May and Gus M. Garmel, “Age plays a significant role. Injuries ranged from 9.7% in those 20 years and younger increasing to 62% in patients 80 years and older (whether or not symptoms were present)”. Hence, life without making arm movements is so miserable, even at a thought. Though there are surgical and non-surgical procedures to help in this regard based on the intensity of tear, early diagnosis through images from X-Rays, Magnetic Resonance Imaging (MRI) or Ultrasound will have a magnificent effect on reducing rotator cuff tears at its preliminary stages. However, physical examination of these images by an expert can never be always effective, owing to the image- capturing noise that hinders the revealing of shape or intensity of the tear. Hence, the modern- day machine learning- based classification approaches can be sought for diagnosis purposes, since they incorporate entities to remove noise and to reveal the presence or absence of the tear in its exact nature. Researchers from Iran and Canada has published in Heliyon a detection approach for rotator cuff tear from MRI images. Collecting about 150 images, they have classified the images to be injurious or healthy using a shallow and quick Convolutional Neural Networks (CNNs). The researchers state that their Computer-aided Diagnosis (CAD) approach could be used as it is in large clinical modules, specifically during athlete evaluation, though the limited access to higher MRI images and processing units emanated as the major limitation. Additionally, the researchers affirm the generalization of the approach to other medical decision problems, encouraging future research to be based on combining CNNs and Long Short-Term Memory (LSTM). Since bone as well as muscle-related ailments are becoming more prevalent in almost all age groups nowadays, new CAD- based technologies are always expected from researchers, whose desire is to serve in the medical field.
Image Courtesy: www.freepik.com
Automating Defect Detection in Fibre-Reinforced Polymers
Inspecting a product for no defects is an important industrial process prior to good product delivery, as no one in the world will be willing to buy a defective product!!! Defects are usually inspected visually through images captured via imaging devices or signals generated from sensors, be it acoustic, thermal or vibration. Nowadays, Non-Destructive Techniques (NDT) for defect detection have been of wider usage for their ability to detect defects like, voids, pores or cracks in the material along with their severity and material strength, without actually causing deformations in the material. This method begins by applying controlled high frequency waves, magnetic fields or infrared waves on to the material surface, which are absorbed and reflected back as sound, magnetic or thermal profiles, respectively. These profiles, on analysis tend to reveal the presence of a defect. Visual analysis of such profiles might not always be accurate and the situation turns worse with numerous materials under testing phase. Machine learning approaches are a boon at this juncture, handling massive data profiles with improved accuracy of defect detection. Advances in Engineering Software, Elsevier, make knows a similar approach by researchers from six countries, whose aim was to inspect the Carbon Fibre- Reinforced Polymer using Barker coded thermography. Here, the material surface was excited with a barker-controlled thermal stimulus, rendering thermal profiles using an infra-red camera. The authors state that the best classification as defective or non-defective polymer was achieved via a One Class Support Vector Machine (OCSVM), when compared to Support Vector Machine (SVM). The Fibre- Reinforced Polymers (FRP) are normally best-known for their increased impact strength, water or corrosion resistance and excellent electrical, mechanical or thermal properties, having great applications in various construction works and industries like, automobile, ship-building, aerospace and rail industries till date. So, developing amenities to automate defect detection in such polymers will definitely be a right choice for structural and mechanical- based researchers.
Image Courtesy: www.freepik.com
Wind Farm Optimization for Increased Energy Production
Do you know that the renewable energy sources have marked the end of the harmful fossil fuel emission-based electric power generation in 2023? Power generation from wind can be quoted as one of the major reasons, since wind never stops during dawn or dusk, though its speed may vary. Usually, using two or three blades connected to a cone-shaped hub, forming the rotor, and a nacelle that contain gearbox, speed brakes and generator mounted over a tower, the wind turbine converts the strong winds from higher altitudes into electricity. However, not all times the speed of wind is same! Whatever may be the speed of wind, stronger winds are experienced upstream or at the front of turbine, while weaker winds are experienced downstream or behind the area crossing the turbine due to energy conversion at the nacelle. The declining wind velocity at the downstream cause wakes, which can interrupt the turbine’s working. But, wind direction perpendicular to the blades only results in maximum power generation and hence, more wakes tend to occur. Hence, a yaw system to change the direction of turbine as per wind direction resides at the intersection of nacelle and tower to simultaneously maximize the output power and minimize the wake turbulence. Controlling the yaw system with rapidly changing wind directions is more challenging. Not only that, the number of wind turbines operating in a wind farm and their optimal placement needs much more consideration to multiply the energy production rates. Researchers from South Korea in Renewable Energy, Elsevier, have presented a joint optimization framework involving Particle Swarm Optimization (PSO) for active control of yaw, as per changing wind velocities, producing an optimal wind farm layout. In accordance to their research, increasing number of wind turbines and smaller size of wind farm layout only yielded the maximum energy efficiency. Since, renewable energy embarks the future of power generation unit, research to increase energy production with little expenditure and less resource usage is always encouraged.
Image Courtesy: www.unsplash.com
Page updated
Google Sites
Report abuse