International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
Deep Learning-Based Multiclass Classification of Diseases in Cucumber Fruit: Enhancing Agriculture Diagnosis
Get PDFAgriculture plays a key role in the economies of many developing nations. cucumber is cultivated vegetable that are grown in large quantities, but the production is regularly affected by diseases, with its yield loss impacted by diseases which include Belly Rot and Pythium Fruit Rot. Early and accurate disease diagnosis is critical for minimizing economic losses and improving crop quality. Traditional method techniques are based on visual identification and time-consuming and often inaccurate, especially for the early stages of the disease. In this work, we aim to tackle these problems and present an automatic cucumber disease classification system by transfer learning. Three convolutional neural network models (pre-trained VGG16, MobileNetV2 and ResNet-50) were retrained on a set of 2400 images containing two disease classes and one normal class. The images were preprocessed with the Contrast Limited Adaptive Histogram Equalization (CLAHE) and background removal by deep learning segmentation to eliminate the background noise and focus only on the informative feature of the image. The models were trained and tested by using training, validation, and test sets with the respective accuracies of 95.28%, 98.06%, and 57.5%. MobileNetV2 showed superior performance to all other models including the highest precision, recall, and F1 score of 0.98, confirming that it was robust and appropriate for real-time disease classification. The results demonstrate that the transfer learning method is conducive to improving the issues of lack of labeled samples and variations in image acquisition and strength, thus providing a reliable model for early disease detection in cucumbers. The system we propose can support farmers and agronomists in early disease management decisions and reduce chemical usage. In the future, we will increase the data set with more disease classes, and develop a mobile APP for field level disease detection
Development of a Machine Learning-Based Predictive System For Classifying Psoriasis
Get PDFPsoriasis is a chronic autoimmune skin condition characterized by inflamed, flaky patches that affect both physical consolation and passionate well-being. Opportune and exact determination is basic for viable treatment; however, it remains troublesome due to its likeness to other dermatological disorders. This research presents a Psoriasis Detection and Severity Classification Framework built on MobileNetV2, a lightweight and effective profound learning demonstrate custom fitted for real-time utilize in resource- constrained situations. Through a basic image-upload interface, healthcare suppliers or patients can yield scalp pictures for robotized investigation. The framework to begin with recognizes the nearness of psoriasis with 90% accuracy, at that point classifies its serious- ness as either “low” or “moderate to severe” with 87% accuracy. This two-step prepare conveys prompt and clinically profitable experiences, supporting more focused on and opportune care. Approved in a clinical setting, the demonstrate illustrates solid unwaver- ing quality and down-to-earth appropriateness. It decreases reliance on expert-driven diagnostics and quickens treatment choices. By coordination AI with restorative hone, this framework improves demonstrative accuracy, streamlines workflows, and engages clini-cians to convey speedier, more personalized care reshaping the scene of dermatological
NeuroSecure-IoMT: Deep Learning Meets Cyber Defense in the Internet of Medical Things
Get PDFIintrusion deduction systems (IDS) are crucial to preserving sensitive medical information from cyber threats. However, issues with multi-class intrusion detection include an imbalanced data set, poor accuracy for minority classes, and a lack of flexibility in handling complex real-world situations. To address these issues, we provide a hybrid framework that combines machine learning and deep learning methods to address these problems. The model uses a random forest classifier for anomaly detection after reducing dimensionality using an autoencoder. The Synthetic Minority Oversampling Technique (SMOTE) was used during processing to ensure equitable class representation and reduce class imbalance. A multi-class intrusion detection dataset tailored to healthcare applications was used to thoroughly test the suggested framework, which provides an impressive 99% accuracy rate. In addition to its excellent accuracy, the model addresses important issues in multi-class Intrusion detection by exhibiting remarkable precision for minority classes and consistent performance across all categories. These results highlight the framework\u27s effectiveness in providing dependable and effective normal detection solutions, which makes it ideal for implementation in crucial sectors like healthcare, their accuracy and data security are crucial
Advancing Diagnosis Capabilities with Smart AI Techniques for Early Symptoms Prediction of Brain Stroke
Get PDFThe brain, a vital organ in the human body, can suffer severe damage during a Brain Stroke (BS) due to blocked blood vessels. The interruption in blood flow and nutrient supply leads to significant symptoms and is considered a medical emergency. BS often results in long-term neurological impairments, complications, or even death, underscoring its critical nature. The World Health Organization (WHO) estimates that BS is the most prevalent cause of disability and death globally. Failure to detect a stroke early may result in delayed treatment, leading to severe complications such as lifelong neurological impairment or death. Early identification with Machine Learning (ML) and Deep Learning (DL) approaches can improve the treatment of patients and reduce the long-term impacts of stroke. The purpose of this research is to predict the signs of a stroke taking place at an early stage employing ML and DL models. To evaluate the efficiency of the approach, a comprehensive training set for BS recognition was collected from a well-known source, Kaggle. The training dataset contains eleven attributes, including age, gender, hypertension, etc., with 5110 records. Multiple classification models, like Support Vector Machine (SVM), Gradient Boosting (XG Boost), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), K Nearest Neighbors (KNNs), and Artificial Neural Network (ANN), were efficiently employed in this study for the identification of initial signs of BS. The suggested ANN has a recognition accuracy of 94.35%, whereas RF has an identification rate of 94.15%. Both have about identical forecast accuracy for BS. The findings of the study revealed that ML and DL approaches have the potential to improve the identification of a variety of illnesses, such as BS, hence reducing the load and subjectivity issues in the medical field that existed owing to earlier traditional methods
Petrochemical Investigation of Secondary Mineralized Volcanogenic Massive Sulfide (VMS) and Supergene Enrichment Economic Deposits in Jandrey-Annar, Upper Dir, Pakistan
Get PDFThis research was about the petrographic and geochemical study of the secondary mineralized Volcanic Massive Sulfide (VMS) deposits of Uthror Volcanics at the Jandrey-Annar study area. Sample examination under the microscope indicates the presence of plagioclase feldspar, sericite, and secondary minerals, such as limonite, hematite, and malachite. Subhedral phenocrysts of pyrrhotite and a highly altered groundmass are indicative of post-magmatic hydrothermal alteration and feldspar sericitization. (Quartz in veins and vugs with undulose extinction indicates recrystallization. The secondary minerals formed by supergene processes were identified by the petrographic index as the products of oxidation and weathering processes of primary sulfide ores. Loss on Ignition (LOI) returns vary from 3.24% to 4.72%, verifying the presence of hydrous mineral species and carbonates, typical for mature secondary mineralized VMS deposits. The rocks are classified as tephrite-basanite, and trachybasalt based on geochemical analysis (AAS and XRF) with the following ranges in their concentrations: SiO₂ (45–48%), Al₂O₃ (16.02–18.63%), CuO (10.48–13.69%), and Fe₂O₃ (5.49–6.20%). The SiO₂ binary plots show positive trends for TiO₂, Al₂O₃, P₂O₅, and K₂O, and negative trends for Fe₂O₃, MgO, CaO, and Na₂O confirming fraction crystallization. High K₂O values indicate the high-K calc-alkaline series. The 10Mn-TiO₂-10P₂O₅ ternary plot classifies the rocks as oceanic island arc basalts, while the R1-R2 plot indicates a late orogenic environment. These results demonstrate mineralization associated with hydrothermal alteration and subduction-related magmatism. Based on analysis of variance (ANOVA) and t-test, high geochemical variation is represented by highly significant (p < 0.01) and significant (p < 0.05) enriched variables including CuO, Fe₂O₃, and MnO, with moderately varying SiO₂, TiO₂, Al₂O₃, and Na₂O, the results indicate hydrothermal alteration and episodic stages of secondary mineralization within the Uthror Volcanics. This high economic potential of the copper ore due to secondary mineralization and supergene enrichment processes has made the region an important target for mineral exploration
Exploring cGANs for Urdu Alphabets and Numerical System Generation
Get PDFUrdu ligatures play a crucial role in text representation and processing, especially in Urdu language applications. While extensive research has been conducted on handwritten characters in various languages, there is still a significant gap in studying raster-based generated images of Urdu characters. This paper presents a generative model designed to produce high-quality samples that closely resemble yet differ from existing datasets. Utilizing the power of Generative Adversarial Networks (GANs), the model is trained on a diverse dataset comprising 40 classes of Urdu alphabets and 20 classes of numerals (both modern and Arabic-style), with each class containing 1,000 augmented images to capture variations. The generator network creates synthetic Urdu character samples based on class conditions, while the discriminator network evaluates their similarity to real datasets. The model’s effectiveness is assessed using key metrics such as the Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Fréchet Inception Distance (FID). The results confirm that the proposed GAN-based approach achieves high fidelity and structural accuracy, making it highly valuable for applications in text digitization and Optical Character Recognition (OCR)
Catalytic Performance of Electro-Oxidative Natural Manganese Sand for Ammonium Nitrogen Removal
Get PDFThe environmental risks associated with ammonium nitrogen (NH₄⁺-N) pollution have led to a growing focus on prevention. Electrochemical advanced oxidation is an effective and eco-friendly method that only requires electricity and electrolytes to remove NH₄⁺-N from wastewater. This study assesses the effectiveness of electro-oxidative natural manganese sand (NMS) in removing ammonium nitrogen under different conditions. Due to NMS’s high redox potential, it significantly enhanced the electrochemical oxidation process, increasing NH₄⁺-N removal and generating reactive chlorine species (ClO⁻/HClO) when NaCl was added. The experiment was also conducted without a catalyst, quartz sand, and natural manganese sand, but NMS removed 86.4% of NH₄⁺-N, outperforming the other treatments. The removal efficiency was tested at five different pH levels (3, 5, 7, 9, and 11), with NMS showing the highest efficiency of 95.2% at pH 9. At a current density of 15.5 mA/cm², the removal rate reached 94.9%, and with a NaCl concentration of 9 g/L, the removal efficiency peaked at 96.2%, driven by increased production of reactive chlorine species (ClO⁻). These results demonstrate the electro-oxidative NMS system as a highly efficient, scalable, and eco-friendly solution for ammonium nitrogen removal in wastewater treatment
Cow Face Detection for Precision Livestock Management using YOLOv8
Get PDFPrecision livestock management is transforming traditional agricultural practices by boosting productivity, increasing yield, and automating tasks, all while reducing labor requirements and minimizing errors. Conventional methods for animal recognition are often unreliable, which has led to a growing preference for using cameras to identify animals, monitor their health, manage data, and maintain cattle records. However, small-scale farms with limited livestock, such as cows and goats, frequently face overfitting problems in traditional machine learning models due to insufficient training data. Identifying individual cows based on facial features becomes more effective after detecting the cow’s face. This study addresses these challenges by fine-tuning YOLOv8, a pretrained model, using a mix of self-captured images and publicly available datasets to detect cow faces in complex environments. Integrating publicly available data and leveraging a pretrained COCO model has significantly improved the model’s ability to generalize and accurately detect cow faces. YOLOv8, equipped with the COCO pretrained model, successfully detects nearly all types of cow faces, which can then be used for individual cow classification. This approach enhances cow recognition accuracy, contributing to more efficient farm management applications
Stereo Vision Based Navigation of Four-Legged Robot Through Unknown Terrain
Get PDFThis research aims to develop a stereo vision-based navigation system for a quadruped robot, enabling it to move autonomously through rough, unfamiliar terrain and detect blockages in sewer pipelines. The robot uses a stereo camera to capture images, which are then processed to create disparity maps and 3D point clouds. These tools help the robot identify and avoid obstacles. Image rectification and 3D mapping are performed using OpenCV, which generates an occupancy grid to distinguish between free and occupied spaces. Based on this grid, the A* algorithm is used to plan the robot\u27s path. To ensure smooth movement, inverse kinematics calculates the required motor angles and applies predefined Bezier curves for stable locomotion
Synergizing Human Behavior and Cybersecurity using Psychometric Scale
Get PDFCybersecurity threats are increasingly shaped by human actions, making it crucial to comprehend the psychological elements that lead to vulnerabilities. This article examines the interplay between human behavior and cybersecurity through the use of psychometric scales to evaluate risk perception, decision-making, and adherence to security protocols. A quantitative research design was employed, using validated psychometric tools like the Human Aspects of Information Security Questionnaire (HAIS-Q) and the Cybersecurity Risk Perception Scale (CRPS). Data was gathered from 200 individuals in different organizational positions and examined using statistical techniques, such as correlation and regression analysis. Findings demonstrated a noteworthy association between psychological characteristics (e.g., risk tolerance, conscientiousness) and cybersecurity practices. People with a greater awareness of risks showed improved compliance with security policies, whereas individuals with lower levels of conscientiousness were more likely to engage in risky online activities. The results indicate that incorporating psychometric evaluations into cybersecurity training can improve threat management by customizing strategies according to personal behavior patterns. This article adds to the expanding research on human-centered cybersecurity strategies, offering empirical data on the impact of psychometrics in enhancing security awareness and compliance. Future studies ought to investigate long-term impacts and cross-cultural assessments of psychometric scales within cybersecurity settings