JOURNAL OF ADVANCED APPLIED SCIENTIFIC RESEARCH

Fast and Efficient Prediction of Honey Adulteration using Hyperspectral Imaging and Machine Learning Models

Mokhtar Al-Awadhi, Ratnadeep Deshmukh — Thu, 30 May 2024 00:00:00 +0000

Recently, honey has become a target of falsification using inexpensive artificial sugar syrup. Current methods for detecting honey adulteration are destructive, slow, and expensive. This paper aims to use hyperspectral imaging (HSI) coupled with Machine Learning (ML) techniques to predict and quantify honey adulteration. The honey adulteration prediction approach proposed in this paper comprises two main steps: spatial and spectral dimensionality reduction and adulteration prediction. We used mathematical averaging to reduce spatial features and employed the Principal Component Analysis and Linear Discriminant Analysis algorithms for spectral feature extraction. Five ML regression models, including Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Decision Tree Regression (DTR), Random Forest Regression (RFR), and K-Nearest Neighbor Regression (KNNR), were used for predicting the sugar concentration in honey. We used a public honey HSI dataset to assess the proposed system's performance. Results show that KNNR outperformed other models in quantifying honey adulteration, achieving a coefficient of determination R2 of 0.94 and a Root Mean Squared Error (RMSE) of 5.12. Findings indicate that HSI coupled with ML models can provide a fast and nondestructive prediction of honey adulteration.

Comparison of Sentences using POS Tagging Tool under Subjective Examination

Aarti P. Raut, C.Namrata Mahender — Thu, 30 May 2024 00:00:00 +0000

The Question answering system is used to generate the correct result that is asked by humans in natural language. In an online examination system, most of the work has been done but still, problems occur in preprocessing i.e. Part Of Speech (POS). POS tagger is used to properly tag each word in the sentences. In this paper, we used two datasets i.e. TREC DATA and data collected from the student. We apply the POS tagger to both datasets and compare the result. For generating the POS tagger we used NLTK and spaCy libraries for comparison. We observed that using those libraries the same word has a different tag. Using both tools, we computed the difference between the words and assigned the count to the POS tagging on that result, we calculate the accuracy of both libraries. The result shows that the spaCy library is best for POS tagging because it generates more correct results as compared to NLTK.

Emotional Speech Recognition for Marathi Language

Bharati Borade, Dr.R.R.Deshmukh — Thu, 30 May 2024 00:00:00 +0000

A spontaneous mental state, emotion does not result from deliberate effort. There are many different kinds of emotions in a speech. Because it enhances interactions between people and technology, automatic emotion identification from human speech is becoming more common today. Several temporal and spectral features of human speech can be extracted. Several methods can be used to categorise pitch-related traits, Mel Frequency Cepstral Coefficients (MFCCs), and speech formants. This study looks at statistical characteristics, including MFCCs and linear discriminant analysis, which were used to categorise these properties (LDA). This article also describes a database of artificially emotionalized Marathi speech. The data samples were collected from Marathi speeches given by men and women that mimicked the emotions that resulted to Marathi utterances that could be utilised in everyday conversation and are interpreted in all analysed emotions. To identify data samples, three essential categories—happy, sad, and angry—were used. For MFCC and LPC, the training accuracy and testing accuracy are 98, 82 and 85,82 respectively.

A Named Entity Recognition System for the Marathi Language

Kadam Vaishali P, C. Namrata Mahender — Thu, 30 May 2024 00:00:00 +0000

Named entity recognition is a complex task in developing many NLP applications. This is one of the essential requirements of language modeling in NLP; without it, it is not possible to proceed further and achieve better results. In this proposed task, we have designed a hybrid technique that is a combination of machine learning and a rule-based approach. This system is to identify such named entities that belong under a specific class, creating a special identification and importance in the meaning generation as well as understanding of the language. This is concerned with the input text. Named entity recognition is important for different group items, such as a person’s name, location or place, animals, organization, time or date, etc. Named entities are informative and good representatives of knowledge. NE also explores the knowledge of artificial intelligence-based systems or expert systems. Using the proposed hybrid model, we have achieved 59.40% performance in identifying named entities and properly labeling for the Marathi

Emotional Intelligence in Text-To-Speech Synthesis in Pali Language Using Fuzzy Logic

Suhas Mache, Siddharth Dabhade — Thu, 30 May 2024 00:00:00 +0000

The field of emotional text-to-speech (TTS) synthesis is making swift progress within the realm of artificial intelligence, holding immense promise to transform our interaction with technology. By using advanced algorithms to analyze and understand the emotional content of text, these systems are able to produce spoken language that accurately conveys the intended emotional tone of the message. Despite the existence of several Text-To-Speech systems across various languages, Pali language is yet to have its own. As a result, we have taken the initiative to create a Text-To-Speech synthesizer exclusively for Pali. Our system offers an end-to-end solution for emotional speech synthesis via Text-To-Speech. We address the problem by incorporating disentangled, well-grained prosody features with global, sentence-level emotion implanting. These well-grained features learn to denote local prosodic differences disentangled from the speaker, tone, and worldwide emotion label. Prosody is usually modeled by rules, so we have implemented the fuzzy logic system to develop a controller for the prosody of Pali speech. The fuzzy controller handles different linguistic parameters in three types of sentences: interrogative, exclamatory, and declarative. The final system produces comprehensible speech that mimics the appropriate intonation for every type of sentence.

In this paper, we introduce and outline the application of a fuzzy paradigm to incorporate a Text-To-Speech system for the Pali language while preserving a rule-based Concatenative synthesizer. In the outline of classic Concatenative TTS systems, we recommend a new method in order to increase Concatenative unit selection computation, directed at increasing synthetic speech perceptual superiority. In order to tackle the problem of phonemes that are prone to multiple descriptions in rule-based speech synthesis, the proposed solution involves a fuzzy system.

In the introductory section, we offer a concise description of the current context surrounding the challenge of emotional speech synthesis. The second section of this paper outlines the notable advancements made in emotional speech synthesis, acknowledging the contributions of various researchers in this field. The third section delves into the technical details of implementing a fuzzy system The last section of the paper presents the main conclusions and future research scope.

Palmprint spoofing detection by using deep learning technique on Multispectral database

Snehal Datwase, Dr.R.R.Deshmukh — Thu, 30 May 2024 00:00:00 +0000

The protection of Biometric systems against attacks is crucial as biometric devices proliferate in the field of personal authentication. The presentation assault is the most prevalent type of attack on biometric systems; it entails presenting a fake copy (artefact) of the true biometric to the sensor in order to gain unauthorised access. The vulnerability in palmprint-based biometric systems has not received much attention despite the substantial threat posed by these assaults. In this research, we show how to detect a spoof palmprint image. Spoofing attacks involving faked images pose a significant threat to biometric systems. For the suggested method, we use the CASIA palmprint database, from which we constructed our own spoof database using printed photos. After that, we did some pre-processing to obtain the ROI image and a noise-free image for feature extraction using the SIFT approach. We use the convolution neural network for classification and the SVM for comparison. We obtained a result of 96.2% for our proposed palmprint system identification and 89% for SVM. But our main goal is to train the model for spoof detection, so we take some normal images and some spoof images for our train model and use the confusion matrix to calculate the accuracy of our model. We obtain an overall accuracy of 86% for our spoof detection by computing the confusion matrix.

Enhancing Diagnostic Accuracy with a Novel Computer Application for Quantitative Analysis of Bio-Medical Infrared Thermal Images

S. Shaikh, R. Manza, P. Yannawar, B. Gawali, N. Shaikh — Thu, 30 May 2024 00:00:00 +0000

Temperature is a parameter that acts as a valuable indicator for understanding persisting disorders and illnesses in the human body. Body surface temperature is measured through the skin and the body’s internal temperature is measured through the mouth or rectum, which are used as vital information reflecting the state of thermo-regulation, a sub-process of the body's homeostasis, which is required for its normal functioning. In a state of functional imbalance, the affected region emits thermal radiation that is above or below the normal range. Thermal imaging of body regions is a beneficial means of detection of such thermal imbalances and the temperature data of each image can be analyzed quantitatively to be able to correlate the results clinically. In this article, a computer-based GUI – MedTherm Image Viewer and Analysis Tool developed in MATLAB is proposed for the processing and quantitative evaluation of thermal images for the purpose of providing supportive aid to the existing medical diagnostic procedures. The suggested graphical user interface (GUI) is beneficial in computing statistical features based on histograms of thermal images that have been recognized in numerous other studies as valuable parameters that assist in clinical diagnostic procedures.

A Machine Learning Approach to Enhance Semantic Understanding in Knowledge Engineering

Bharat A. Shelke, Vikas T. Humbe, C. Namrata Mahender — Thu, 30 May 2024 00:00:00 +0000

Developing complex systems in environments of various domains need effective way to share, capture, and integrate knowledge of experts. “Modern Knowledge Engineering (KE)” systems meet this function to execute dignified knowledge with highly dedicated languages and environments. However, commitment of such environments to their application domain poses restrictions on incorporation of KE across the domain. Using Semantic Understanding (SU) can deliver a domain-neutral option to formalize knowledge and integrate data to reduce the effort needed for integration of knowledge of various domains in one representation.

This paper discusses machine learning approaches used to solve problems related to knowledge engineering. Semantic Understanding has seen a lot of improvements over the decades as per industrial demands and human needs. This new era is related to teaching machine to learn itself and understand the purpose and concept of its use with algorithms. This paper discusses semantic technology used in machine learning and its idea. It briefly discusses the important role of “machine learning and semantic technology”.

Transfer Learning Using Teachable Machine For Classification Of Glassware In Chemistry Lab

Trupti Satpute, Dr. R. V. Kulkarni, Prakash Bansode — Thu, 30 May 2024 00:00:00 +0000

Image classification is an important use case of deep learning algorithms. Convolution Nural networks, CNNs, have evolved to an extent where pretrained models can be used to train new models. The technique used for this type of model building activity is called as Transfer lerning. We have developed an image classification model using transfer lerning to classify lab glassware used in Chemistry lab. This model can be used for training purpose for the students in high schools who are not much aware about the practical implementation of laboratory experiments. We have used subset of Labpics dataset developed by Eppel et.al. We have used teachable machine as a platform to build this model with very limited computational resource. With transfer learning mechanism used by teachable machine platform we were able to achieve ~83% accurate image classification model

Text-to-Speech Synthesis for Hindi Language Using MFCC and LPC Feature Extraction Techniques

Shaikh Naziya Sultana, Ratnadeep R. Deshmukh — Thu, 30 May 2024 00:00:00 +0000

India is a large country with over a billion populations who speak numerous languages. 43% of Indians speak Devanagari Hindi script, followed by Bengali, Telugu, Marathi, and other languages. The widespread generation of content and accessibility would therefore greatly benefit from text-to-speech systems for such languages. In this research work we improve the already available Text-to-Speech (TTS) system using advance preprocessing techniques to the Hindi corpus database and applied various feature extraction techniques for better result. Finally we got the accuracy as 98% using MFCC and LPC feature extraction techniques. The developed model is capable for getting the input from audio file and read it loudly using developed TTS system.

Vision-based primary localization method for SLAM mobile robots

Mingcen Gu, Stephen Karungaru, Kenji Terada, Yuehua Li — Thu, 30 May 2024 00:00:00 +0000

The AMCL (Adaptive Monte Carlo Localization) algorithm with visual provision of initial values is proposed to address the slow localization speed caused by conventional laser SLAM (Simultaneous Localization and Mapping) without initial poses and the global localization failure after a robot abduction event. In the initial map building phase, the ORB (Oriented FAST and Rotated BRIEF) feature values are extracted from the camera and the wall corners are identified, and then the pose information is stored in the database and a feature dictionary is constructed. After restarting, the dictionary is called to perform loopback detection by receiving the images captured by the current camera, and a successful detection results in a rough initial pose. If the detection fails, the initial pose is roughly calculated by identifying the wall corners. Finally, the particle filtering algorithm scatters particles in a small area near the obtained pose and converges to obtain a relatively accurate pose.

Image-based Pretreatment Study of Rice Blast Disease

Thu, 30 May 2024 00:00:00 +0000

Rice blight has a great impact on rice yield and can lead to yield reduction of up to 70% in severe cases. Traditional detection methods require professional technicians to operate and are costly and inefficient, and cannot detect rice diseases in real-time. In this paper, we applied image detection technology to study rice blast disease based on the Matlab platform. Firstly, a basic rice blast database is built, and then a discussion is made on how to effectively improve the recognition success rate of rice blast images by two aspects: image pre-processing and feature extraction. The main research contents are as follows. (1) After studying the existing plant disease database, a basic rice blast database was constructed by field photography and other means. (2) Preprocessing of the collected rice blight images. Using the algorithm of rgb2gray function in Matlab, the images were grayed out; based on this, median filtering was used for noise reduction; then histogram equalization technique was used for image enhancement to increase the contrast and make the images clear; finally, various segmentation algorithms were used for image segmentation. (3) For the pre-processed rice blight images, feature extraction was performed in terms of the color of the disease to pave the way for feature selection.

Pedestrian detection algorithm based on improved YOLOv3

Meiqing Wang, Stephen Karungaru, Terada Kenji — Thu, 30 May 2024 00:00:00 +0000

The ability to accurately detect pedestrians in the area of interest in real time is crucial in the field of autonomous driving. An improved YOLOv3 model is proposed for pedestrian detection. Firstly, a lightweight model that incorporates a residual network module approach and a CBAM attention mechanism is added to the structure to enhance the feature representation capability of the network. Experimental results show that the improved YOLOv3 target detection model raises the detection accuracy by 4% compared to the original algorithm, and the accuracy precision is improved to a large extent, which verifies the feasibility and effectiveness of the improved YOLOv3 model for pedestrian detection.

Indoors Fitness Training Monitoring based on OpenPose

J.Haoran, S. Karungaru, K. Terada — Thu, 30 May 2024 00:00:00 +0000

With the continuation of the COVID-19 pandemic, people's daily life has changed. The changing life habits are reflected in the increasing number of hours working at home. Mostly affected is physical fitness, because of limitations or fear of the gym/outdoors or effective exercise indoors. However, with the arrival of the post-pandemic era, although working at home has improved, the fitness problem still haunts people. Some people have become accustomed to home fitness and are no longer limited to the traditional gym or gymnasium. However, proper and safe exercising is still a challenge due to the lack of live coaching. With the advent of artificial intelligence and the improvement of virtual reality (VR) and augmented reality (AR) capabilities, the options for live off-site coaching have become feasible. This study is based on OpenPose technology in artificial intelligence to monitor the standard of people's movements in-home fitness. The study results are encouraging.

Deep Learning Approach towards Emotion Recognition Based on Speech

Thu, 30 May 2024 00:00:00 +0000

Feelings are incredibly vital in the internal actuality of humans. It's a means of communicating one's point of view or emotional condition to others [5]. The birth of the speaker's emotional state from his or her speech signal is appertained to as Speech Emotion Recognition (SER) [2]. There are a many universal feelings that any intelligent system with finite processing coffers can be trained to honour or synthesize as demanded, including Neutral, wrathfulness, Happiness, and Sadness. Because both spectral and prosodic traits contain emotional information, they're utilized in this study for speech emotion identification. One of the spectral parcels is Mel- frequency cepstral portions (MFCC). Prosodic variables similar as abecedarian frequency, loudness, pitch, and speech intensity, as well as glottal factors, are utilized to model colorful feelings. For the computational mapping between feelings and speech patterns, possible features are recaptured from each utterance. The named features can be used to identify pitch, which can also be used to classify gender. In this study, the gender is classified using a Support Vector Machine (SVM) on Ravdess dataset. The Radial Base Function and Back Propagation Network are used to honour feelings grounded on specified features, and it has been shown that the radial base function produces more accurate results for emotion recognition than the reverse propagation network.

Deep learning models for stock prediction on diverse datasets

Rachna Sable, Shivani Goel, Pradeep Chatterjee, Mani Jindal — Thu, 30 May 2024 00:00:00 +0000

Market forecasting has attracted the interest of investors all over the world. The investors are looking for an accurate and reliable forecasting model that can fully embrace the extremely volatile and nonlinear market behavior. It is now possible to design effective stock price prediction algorithms due to the abundance of data, the quick advancement of AI and machine learning techniques, and the machine's increased computational capability. Deep learning algorithms are particularly successful in modelling market volatility. To forecast the closing prices of three stocks: Apple (AAPL), Google (GOOG), and Amazon (AMZN), Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) are implemented and compared. The stock data was obtained from yahoo finance for one year, three years and five years. The Root Mean Square Error (RMSE) metric and loss are employed for evaluating the model’s performance.

Self-Governing Feedback Network (SGFN) Based Super Resolution for bean leaf disease detection

P.V.Yeswanth, M.Saisanika, S.Deivalakshmi — Thu, 30 May 2024 00:00:00 +0000

Crop loss caused by diseases that result from a range of insects, bacteria, viruses, and fungi has been a severe concern for generations that demands global attention. As a result, diagnosing crop diseases as soon as feasible can dramatically reduce production loss and enhance monetary value. The Self-governing Feedback Network (SGFN) model is suggested in this paper for producing Super Resolution images from low-resolution bean leaf images and recognizing disease. On the bean leaf dataset, the proposed SGFN model is tested for super-resolution factors 2, 4, and 6. PSNRs of 31.27, 35.653, and 37.721 are achieved for super-resolution factors 2, 4, and 6, respectively, with classification accuracies of 99.54, 98.73, and 97.64.