Journal of Advancement in Electronics Signal Processing

Intelligent Rover for Real-time Coconut Disease Detection and Precision Fertilizer Application

2026-06-06T12:18:56+00:00

Modern agriculture faces significant challenges in early disease detection and efficient resource management. This study presents an autonomous mobile platform integrating computer vision and deep learning for real-time coconut tree health assessment. The proposed system employs the YOLOv8 architecture for disease identification and growth stage classification, achieving detection rates exceeding 90% in field trials. A Raspberry Pi-based processing unit coordinates with Arduino microcontrollers to enable remote operation through a web-based interface. The platform captures live imagery, performs on-device inference, and provides fertilizer recommendations based on detected conditions. Field validation demonstrates the system’s capability to reduce manual inspection time by 75% while maintaining detection accuracy comparable to expert assessment. The integration of autonomous navigation with precision agriculture techniques offers a scalable solution for plantation monitoring and targeted intervention. The rover system identifies four critical disease classes, including Bud Rot, Stem Bleeding, Grey Leaf Spot, and Bud Dropping, with an overall accuracy of 93.1%. Real-time processing operates at 26 frames per second with minimal latency, enabling smooth video streaming to a Flutter-based mobile application. The system’s distributed architecture separates high-level AI processing on Raspberry Pi from time-critical motor control on Arduino, ensuring reliable operation. Battery-powered operation provides 4.5 hours of continuous monitoring with WiFi connectivity extending up to 50 meters. The automated fertilizer recommendation engine achieves 100% alignment with expert agricultural protocols, supporting sustainable farming practices through optimized chemical application. This cost-effective implementation using commercially available components makes precision agriculture technology accessible to small and medium-scale farmers, addressing critical labor shortages while improving crop health management.

JARVIS: A Speaker-Authenticated Voice Command System Using GMM-SVM Fusion and NLP

2026-06-19T10:48:55+00:00

This study presents the development of JARVIS (Just A Rather Very Intelligent System), a secure voice assistant designed for hands-free interaction and reliable user authentication. The system combines Support Vector Machine (SVM) and Gaussian Mixture Model (GMM) techniques to verify speakers with improved accuracy and protection against unauthorized access. To improve audio clarity, preprocessing methods such as noise reduction, silence removal, pre-emphasis filtering, and normalization are applied. A detailed feature set including MFCC, delta coefficients, chroma features, and spectral contrast is used to capture unique voice patterns. For command understanding, the system adopts a three-level intent recognition approach using TF-IDF-based Logistic Regression, keyword detection, and fuzzy matching. JARVIS is implemented as an interactive Streamlit application with features like waveform visualization, MFCC heatmaps, pitch tracking, emotion analysis, and command history monitoring. Experimental evaluation shows that the system delivers reliable authentication, quick response time, and effective multilingual support for English, Hindi, Kannada, Tamil, and Telugu, making it suitable for secure smart automation and voice-controlled applications.

Smart Brain Tumor Detection and Severity Monitoring using Deep Neural Networks

2026-07-04T04:44:48+00:00

Rapid and accurate diagnosis of brain tumors is essential for planning appropriate treatment and improving patient survival. This paper presents a DL-based automated system as a means of categorizing brain tumors and staging using Magnetic Resonance Imaging (MRI) data implemented in MATLAB. The proposed approach employs a transfer learning strategy using a pre-trained ResNet-50 convolutional neural network to extract discriminative characteristics from brain MRI images. Before grouping, comprehensive pre-processing methods such as scaling, normalization, denoising, and data augmentation are used. The technique estimates tumor severity levels while categorizing MRI images into meningioma, glioma, pituitary tumor, or normal categories. Standard measures such as F1-score, confusion matrix, accuracy, precision, recall, and Receiver Operating Characteristic (ROC) analysis are used to assess performance. Experiments show that the suggested model outperforms traditional techniques and delivers dependable staging performance and high classification accuracy, confirming its applicability for computer-aided clinical decision support systems.