International Journal of Artificial Intelligence, Machine Learning and Intelligent Systems

Autonomous Vision-Based Fire Detection in Rice Farms: A Solar-powered IoT System Utilizing Hybrid CNN LSTM Networks

Tue, 21 Jul 2026 00:00:00 +0000

Agricultural fires in rice paddies that burn out of control are a major hazard to agricultural production and rural lives globally. Standard fire detection systems, based on point sensors, smoke and heat, ineffectively detect fire in open field conditions where wind currents easily carry signals away. This work suggests a solar energy self-governing Vision-Based Fire Detection (VBFD) system designed for an environment of remote monitoring of a rice farm. This system has a hybrid spatial-temporal architecture. To eliminate the effect of the environment, a CNN is used to extract the spatial features from the images by performing the convolution task, and to distinguish image details, an LSTM algorithm is used to perform the temporal (temporal “flicker”) feature extraction task. The system runs on a “Raspberry Pi 4B” edge computing platform, which uses on-device inference to eliminate latency and reliance on the cloud. The study evaluated a custom-trained CNN using a validation set of 199 images and compared it to a pretrained CNN with a MobileNetV2 backbone. Experimental results show that the custom CNN outperforms the MobileNetV2 model with a better overall correctness of 97% and a fire detection recall of 0.98 in comparison to the MobileNetV2 with 93% accuracy and 0.92 recall. The custom model’s performance in identifying a fire with a high probability when it results in a false negative only 2% of the time is crucial to minimizing the effects of catastrophic losses. It also has an integrated SIM800L GSM module that allows immediate SMS notification sending and an integrated solar-powered UPS system for 24/7 off-grid operation. The study delivers a scalable, high-recall, early warning solution that can play a substantial role in current Climate Smart Agriculture. The present study can be perceived as an effective, cost-effective tool for ensuring safe food production in high-risk areas.

RoadCare AI: Smart Road Health Monitoring and Maintenance Framework Using Deep Learning and YOLOv12

Asha Shantappa, Fariya Noorain — Mon, 22 Jun 2026 00:00:00 +0000

Road infrastructure degradation has emerged as a critical challenge affecting transportation safety and economic efficiency, where the increasing presence of cracks and potholes contributes to accidents, vehicle damage, and delayed maintenance response. Within the paradigm of intelligent transportation systems, the RoadCare AI system is conceptualized as a deep learning-driven framework that enables automated detection of road damage through real-time image and video analysis. Traditional inspection methods rely on manual observation, which introduces delays, inconsistencies, and increased operational costs, thereby limiting effective infrastructure management. The proposed system integrates computer vision-based preprocessing with advanced deep learning models to enhance detection accuracy under diverse environmental conditions. The framework utilizes techniques such as noise reduction, contrast improvement, and normalization to ensure consistent input quality while enabling efficient feature learning. The enhanced data is processed through ResNet for feature extraction, which captures complex spatial patterns associated with road damage. Furthermore, the system employs YOLOv12 for high-speed object detection, enabling precise localization of cracks, potholes, and manholes in real-time scenarios. The integration of detection and alert mechanisms facilitates immediate reporting of road hazards, thereby supporting timely maintenance actions. Experimental evaluation demonstrates that the proposed YOLOv12 C3ECA DSA model achieves approximately 97% mAP with a training loss of 0.06 and validation loss of 0.05, confirming high accuracy and robust performance while maintaining computational efficiency. The system establishes an effective solution for automated road damage detection, contributing to improved road safety and intelligent urban management.

A Systematic Review of Research Gaps and Future Directions for Generalizable Autonomous Systems for Causal and Neuro-Symbolic Artificial Intelligence

S. Hassain, Joy Kumar — Wed, 15 Jul 2026 00:00:00 +0000

Autonomous artificial intelligence systems continue to rely heavily on correlation-based learning mechanisms. The limited application of causal learning mechanisms has resulted in poor causal reasoning, low logical consistency, low adaptability, and low transparency in autonomous decision formation. This systematic review aims to critically evaluate existing research with respect to the incorporation of autonomous artificial intelligence models with causal representation learning mechanisms for effective decision formation, neural learning with symbolic reasoning for logical consistency, and generalization with respect to unseen operational scenarios. The quantitative synthesis has revealed that fewer than 30% of artificial intelligence models incorporate causal modelling mechanisms, while integrated neuro-symbolic models are limited to 25% of intelligent system frameworks. In addition, almost 60% of frameworks are found to be devoid of continuous self-learning mechanisms with respect to adaptive feedback, while more than 65% are found to be devoid of interpretable reasoning chains with respect to autonomous decision formation. The gaps with respect to autonomous artificial intelligence frameworks highlight the need for integrated frameworks with respect to unified causal-neuro-symbolic models with self-learning, explainable, and generalizable characteristics.