Journal of Artificial Neural Networks and Learning System (p-ISSN: 3049-0758, e-ISSN: 3048-6629)

KrishiMitra – Development of a Hybrid Deep-Learning Algorithm for Detection of Diseases in Tree Leaves

2026-01-30T09:13:37+00:00

Plant diseases pose a major challenge to agricultural productivity in developing economies, where smallholder farmers often lack access to timely diagnostic support and expert guidance. Early disease identification through leaf image analysis offers an effective pathway for enabling timely intervention and reducing crop losses. This paper presents KrishiMitra, an Internet of Things (IoT)- enabled smart agriculture framework designed for practical, cost-effective plant leaf disease detection. The proposed system explores a hybrid deep learning architecture that integrates Convolutional Neural Networks for local feature extraction with attention-based mechanisms for enhanced feature representation, while transformer-based components are incorporated at the architectural level to support future contextual learning enhancements. The study focuses on system design, model development, and preprocessing strategies, supported by experiments conducted using the Plant Village dataset augmented to reflect real-world agricultural image variations. KrishiMitra is architected with edge deployment as a primary design objective, targeting low-cost devices such as the ESP32-CAM to enable decentralized, connectivity-independent operation. A web-based platform is developed to support image acquisition, disease prediction, and result visualization for farmer accessibility. While full-scale hardware optimization and extended field validation are ongoing, the presented work demonstrates the feasibility of deploying hybrid deep learning techniques within resource-constrained agricultural environments. This research establishes a foundation for scalable, farmer-centric disease detection systems and outlines clear directions for future enhancements, including edge-device optimization, planned explainable AI integration, and full-scale system deployment.

A Hybrid Deep Learning Framework with Complementary Feature Fusion for Automated Multi-Class Waste Sorting

2026-02-24T05:37:11+00:00

Sorting of the waste using automated methods of control is also an essential part of the contemporary system of waste management since it allows sorting of the material correctly and enables sustainable recycling. The paper suggests a novel hybrid deep learning framework known as multi-class waste classification that will combine the features and capabilities of Swin-Tiny Transformer and ConvNeXt-Tiny networks in the form of complementary feature learning. The proposed framework will combine both network deep feature representations at the feature level to create a single feature, as well as classify it, where the overall accuracy is 97.85%. The model achieves macro-averaged precision, recall, and F1-score of 97.37%, 97.09%, and 97.21%, respectively, and the weighted precision, recall, and F1-score of 97.87%, 97.85%, 97.85%, respectively, which shows no drift in the model and balanced performance on all the classes. The evaluation of the framework is based on a publicly accessible waste dataset collection of 15,515 images of 12 types of classes, which guarantees assessment of multiple classes. The efficacy and resilience of the suggested hybrid model have been confirmed through the use of experimental results, which show that the hybrid model is always more effective and stronger in comparison to the individual architectures.

Restoring and Processing Images by Applying Geometric Transformations with Respect to A Reference Image

2026-02-28T10:31:48+00:00

Imaging systems are widely used in applications such as commercial photography, microscopy, aerial imaging, astronomy, and space exploration. However, the acquired images or videos often suffer from blur caused by lens imperfections, transmission media, image processing algorithms, or motion of the camera or subject. Quantifying and mitigating this blur is a critical challenge. Image processing and restoration techniques aim to enhance image or video quality using various approaches, primarily through the manipulation of pixel intensities. An image can be mathematically modeled as a two-dimensional function f(x,y)f(x, y)f(x,y), where xxx and yyy represent spatial coordinates. In this study, geometric transformations are applied for image restoration and processing. The methodology involves reading the image into the algorithm, defining pixel coordinates, applying a geometric transformation with a rotation angle of 31 degrees, performing inlier–outlier matching, and generating the restored image. Additionally, performance is evaluated using mean squared error (MSE) and peak signal-to-noise ratio (PSNR). The entire investigation is implemented using a MATLAB .m script.

An Exploring Artificial Intelligence in Modern Image Creation

2026-03-19T12:34:43+00:00

Artificial Intelligence (AI) has changed the way images are created. Today, AI can generate high-quality, realistic, and creative pictures on its own. Methods like GANs, diffusion models, and neural style transfer help AI create new artwork, improve photos, and change images into different styles. These technologies make the work easier and faster for people in fields like advertising, entertainment, gaming, and graphic design. AI tools allow both professionals and beginners to create beautiful images, making art and design more accessible to everyone.
However, using AI in image creation also brings some challenges. There are concerns about copyright, originality, and the risk of creating fake or harmful content. AI-generated images also raise questions about what creativity means and how much value human artists bring to the process. As AI continues to grow, it is important to use it responsibly encouraging innovation while also respecting ethics and society. This project explains the main technologies used in AI image creation, how they work, where they are used, their advantages, and the challenges they bring. It shows how AI is changing the way people create images and helping them be more creative. It also talks about important topics like ethics, copyright, and the responsible use of AI-generated images.

Prompt Engineering Techniques in Large Language Models: A Study

2026-04-01T11:04:37+00:00

Large Language Models (LLMs) are revolutionizing the field of Artificial Intelligence by empowering computers to produce text that is comparable to natural language, make logical decisions, write coding scripts, and help make complex decisions. The effectiveness of these models is largely dependent on the structure, clarity, and context of the prompts given to them. As a result, prompt engineering has evolved as a powerful paradigm to effectively direct LLMs towards producing accurate, informative, and context-driven text. This study undertakes a detailed exploration of prominent prompt engineering techniques, namely Zero-Shot Prompting, Few-Shot Prompting, Chain of Thought Prompting, Role-Based Prompting, Instruction Tuning, Self-Consistency, and Prompt Chaining. The study undertakes a detailed analysis of the underlying principles of each of these techniques, their pros and cons, as well as their suitability for various domains of application. A comparative evaluation is carried out to assess the trade-offs of each of these techniques based on their reasoning power, computation costs, accuracy of responses, and implementation complexities. Additionally, challenges such as prompt sensitivity, hallucinations, and bias are discussed to further emphasize the importance of reliability and ethics within LLMs. The paper further explores new research avenues that are being pursued within the field of LLMs, including prompt optimization through automation and multimodal AI systems. The study shows that structured and strategic prompt engineering is essential for improved reasoning capabilities of LLMs, further reinforcing the importance of this component within AI systems.