The Internet of Drones (IoD) has become increasingly important in applications such as forest inventory, leveraging advanced sensors and internet connectivity to enable efficient data collection. Compared to traditional methods, IoD offers superior cost-effectiveness. However, its reliance on public channels, unreliable connectivity, and dynamic environments poses significant security and privacy challenges. Safeguarding forest inventory data is essential to maintaining accuracy, preventing unauthorized access, and mitigating the risk of data manipulation, which can lead to suboptimal management decisions. To address these concerns, it is essential to design a lightweight authentication protocol that secures IoD communication, accounts for network bandwidth limitations and scalability, and supports integration with emerging technologies. This manuscript introduces a new Authentication and Key Agreement (AKA) protocol specifically designed for the Internet of Drones (IoD), leveraging asymmetric cryptography and aggregate signatures to enhance security and privacy in forest inventories with fog computing. Its robustness was confirmed through informal and formal security analyses by the AVISPA tool and the ROR model, demonstrating resistance to known attacks and superior communication, computational, and energy performance compared to existing protocols.

The Coffee Berry Borer is the most destructive pest affecting global production of Coffea arabica. Early detection of pest-induced fruit damage remains challenging due to the small size of infestation symptoms and the dense clustering of coffee berries under complex field conditions. This study evaluates optimized object detection architectures designed to improve the balance between detection accuracy and computational efficiency. Three baselines were established: YOLOv8n (M0), YOLOv11n (M1), and YOLOv26n (M2). Seven architectural variants (M3–M9) were then developed by integrating FasterNet, SimSPPF, and EMA. Experimental results showed that M0 achieved the highest detection accuracy (mAP@0.5 = 0.9534 and 6.09 GFLOPs), whereas model M6, combining FasterNet and SimSPPF, provided the best accuracy–efficiency trade off with mAP@0.5 = 0.9446 and 5.12 GFLOPs. Pareto analysis confirmed M6 as the optimal configuration. Finally, in situ validation across 25 points achieved a mean F1-score of 0.7255 (SD = 0.0504) for infected berries despite cast shadows, proving its readiness for real-time agricultural deployment.

Underwater optical sensor networks are essential for fish monitoring, yet imagery is often affected by illumination variability, low contrast, and complex backgrounds. Attention mechanisms are vital for feature representation in deep networks, yet existing approaches often struggle with spatial information loss and limited multi-scale interaction under such challenging conditions. This paper introduces Convolution to Interactive Capture and Recalibration Enhancement (C2ICARE), a lightweight attention module designed to overcome these challenges. The principal contribution of C2ICARE is the adaptation of memory interaction principles into an edge-oriented attention framework that enhances feature discrimination while maintaining computational efficiency. The architecture employs three core innovations: a 1:3 memory-feature split to preserve context while reducing cost, parallel multi-scale depthwise convolutions (3 × 3 and 7 × 7) for fine-grained and broad feature extraction, and a cross-branch interaction mechanism coupled with a ConvNeXt-style feed-forward network that avoids dimensionality reduction. Experimental results on an underwater fish dataset demonstrate that YOLO26n with C2ICARE achieves a mean average precision (mAP@0.5:0.95) of 0.7033, outperforming Coordinate Attention (+3.8%), FasterBlock (+1.7%), and CBAM (+0.4%) while adding only 0.05M parameters and 0.16 GFLOPs. Multi-objective Pareto Frontier analysis confirms that C2ICARE provides an effective balance between accuracy, efficiency, and generalization for resource-constrained deployment. EigenCAM visualizations further validate that the model focuses on biological morphology rather than background noise. Its lightweight design enables seamless integration with underwater sensor networks and fog platforms for real-time fish detection in aquaculture, commercial fisheries, and scientific research. Future work will investigate broader marine applications and cross-platform deployment scenarios. The code is available on GitHub.

Wireless power transmission (WPT) is a key enabling technology for autonomous Internet of Things (IoT) sensor networks deployed in remote or hard-to-access environments. This paper presents an intelligent resonant wireless energy distribution system that integrates power transfer, sensing, communication, and adaptive control within a unified Wireless Sensor Network (WSN) architecture. The proposed system introduces multi-role nodes capable of operating as transmitters, relays, or receivers, enabling scalable and battery-free network deployment. A parameterized analytical and simulation framework is developed to model power transfer behavior, and a functional prototype is implemented and experimentally validated. The system incorporates a rule-based adaptive control mechanism and IoT communication (MQTT/HTTP) for real-time monitoring and dynamic power regulation. Experimental results demonstrate wireless energy transfer over distances up to 35 meters under controlled conditions, supporting continuous operation of distributed sensor nodes. The proposed approach advances WPT-enabled IoT systems by providing a scalable, energy-autonomous architecture for distributed sensing applications.

Volatile organic compounds (VOCs) are central to the aromatic and therapeutic properties of essential oils (EOs), with their profiles serving as reliable indicators of EO quality. In Cistus ladanifer, the synthesis and emission of VOCs—particularly terpenic hydrocarbons—are strongly influenced by environmental variables such as temperature, humidity, and solar radiation. Traditional EO quality assessment methods, including gas chromatography (GC), although highly accurate, are costly, labor-intensive, and destructive. This study proposes a smart sensor system that utilizes a low-cost array of MQ gas sensors combined with machine learning (ML) algorithms for non-destructive classification of VOC fingerprints in Cistus ladanifer EO. VOC data from 33 EO samples were collected using MQ sensors and paired with environmental datasets (Daily, 15-Day, and All) obtained from weather stations in the cultivation areas. Gas chromatography coupled with flame ionization detection and mass spectrometry (GC-FID/MS) was used as the reference method to quantify the concentrations of terpenic hydrocarbons. A total of 5,154 data points were used, with 75

ABSTRACT In an era where sustainable agriculture is imperative, precision crop management emerges as a vital strategy to enhance yield while conserving resources. The agricultural sector, increasingly shaped by advancements in Artificial Intelligence (AI), is progressively adopting gas sensors—particularly metal‐oxide semiconductor (MOS) types—as critical tools for data‐driven monitoring and decision‐making. Despite their potential, widespread use of gas sensors in agriculture remains constrained by high initial costs, limited user training, complex data requirements, and uncertain returns. This study highlights the transformative role of cost‐effective MOS gas sensors in early disease detection, yield enhancement, and system efficiency. Readily available from global retailers such as AliExpress, Amazon, and eBay—some priced as low as $0.99—these sensors offer promising accessibility. However, challenges persist in VOC quantification, power consumption, selectivity, durability, and signal stability. The study explores these limitations alongside proposed solutions and research directions. AI methodologies such as Support Vector Machines (SVM), Partial Least Squares Discriminant Analysis (PLS‐DA), and Artificial Neural Networks (ANNs) show potential to improve selectivity and reduce drift error, contingent on access to large, labeled datasets. As technological refinement progresses, MOS gas sensors are poised to play an expanding role in precision agriculture, aligning environmental management with data‐centric innovation.

Timely detection of abiotic stress is critical for precision agriculture, enabling early intervention and sustainable crop management. This study introduces a novel, nondestructive sensor system for real-time detection of salinity stress in hydroponically grown arugula [Eruca sativa (Mill.) Thell.] using low-cost metal-oxide-semiconductor (MOS) gas sensors. Arugula plants were cultivated under controlled greenhouse conditions and exposed to three salinity levels [0-, 40-, and 80-mM sodium chloride (NaCl)]. Volatile organic compounds (VOCs) emissions from plants were captured using a dome-based enclosure equipped with MQ2, MQ135, and MQ137 sensors and continuously recorded over eight days. Sensor outputs revealed distinct VOC patterns associated with increasing salinity stress, validated through physiomorphological measurements. A machine learning (ML) pipeline-comprising K-nearest neighbors (KNNs), support vector machine (SVM), and random forest (RF) classifiers-was trained on the VOC data, achieving up to 99.15% accuracy in identifying stress levels. Configurations using the full sensor array consistently outperformed single- and dual-sensor models, both in terms of classification performance and confidence. The system employed wireless sensor network (WSN) architecture to enable scalable and distributed environmental monitoring. Each node integrated an Arduino Mega 2560 for analog signal acquisition and was connected to a Raspberry Pi 4, serving as a gateway for the WSN. Data were then structured and stored in a MariaDB database. This work is the first to demonstrate the use of low-cost gas sensors for VOC-based salt stress detection in arugula, offering a promising tool for early stress diagnosis in controlled-environment agriculture. The approach provides a scalable, real-time solution to enhance crop monitoring, with potential applications across a wide range of crops and stress conditions.

Salinity stress severely constrains crop productivity, creating an urgent need for rapid, scalable, and non-destructive diagnostic tools compatible with intelligent farming systems. This study presents an integrated phenotyping and analytics framework that combines RGB imaging, machine learning, and edge enabled wireless sensor network for accurate detection of three levels of salinity stress, 0 mM, 100 mM, and 200 mM of NaCl, in peppermint. Daily lateral and nadir images were acquired over a ten–day period and subjected to robust color space segmentation, after which two complementary feature extraction strategies were applied, namely RGB–based and histogram–based representations, with vegetation indices computed for both feature sets alongside morphological descriptors. Machine learning models were evaluated with emphasis on predictive accuracy, computational efficiency, and suitability for deployment on resource constrained devices. Nadir image features consistently outperformed lateral perspectives, achieving classification accuracies of up to 98.30

Monitoring natural environments is currently a hot research topic, as the early detection of an anomaly or the presence of contaminants in such a setting allows for mitigating environmental damage. In this regard, these tasks have been carried out using nodes that executed code sequentially, mainly due to their inability to execute multiple threads. However, in this article, we propose the collaborative use of sensor nodes within a network, so that a single node can be multipurpose and provide service by monitoring different sets of variables for the same environment. To this end, we propose using autonomic computing, a self-management mode for nodes that allows them to interact with each other and execute different codes as needed. After testing the system, it has been confirmed that this type of operation generates significant energy savings for each node in the network and that, ultimately, it is possible to work with a smaller number of nodes, which implies consequent economic savings.

This study introduces an innovative Edge Computing Wireless Sensor Network and Designing a new algorithm for diagnosing orange fruit diseases. The network combines Raspberry Pi using wireless technologies like Zigbee and LoRa with Wireless Mesh Routers using Wireless Technologies like LoRa and Cellular technologies. By using a new system that includes a YOLOv8 model and an image processing algorithm that detects the color spectrum of the diseased part of the fruit, it is possible to quickly identify certain diseases, such as canker, black spot, and melanosis. The system achieves a high accuracy of 92.2% in disease detection. This cost-effective and efficient solution offers farmers a practical tool for early disease detection, enabling timely interventions to protect crops and improve overall agricultural outcomes. In this study, in connection with the proposed algorithm, 97 images of diseased orange fruit, including Canker, melanosis, and black spot, as well as healthy oranges have been tested. It has also been tested in an orange orchard. The proposed new model successfully identified orange black spot disease with 30 correct detections out of 32 images and 2 errors, melanosis disease with 18 correct detections out of 21 images and 3 errors, canker disease with 9 correct detections out of 11 images and 2 errors, and 33 images of healthy oranges fruits with 100% accuracy. The Python codes for the proposed model and the dataset used in this study are available in a GitHub repository and accessible to the public.

Water quality is being altered by human activities and monitoring them is highly recommended. Nevertheless, low-cost monitoring nodes are generally limited in terms of computation capacity. In this paper, we analyse the performance of different machine learning algorithms trained and tested in nodes as part of edge computing in a water quality monitoring network. The performance of five well-known algorithms for data classification in multiclass and binary classification is assessed. The algorithms are coded to run in a low-cost node, and their performance is compared with Matlab results. Accuracy and precision are the metrics used for the comparison, with the artificial neural network as the algorithm with higher performance, followed by the Naive Bayesian classifier. The accuracy for the multiclass classification in the node is slightly better, 98.66

The growing need for sustainable and renewable energy sources has become critical with the Internet of Things (IoT) advancement. IoT relies on low-power, battery-operated devices, but the limited lifespan of these batteries requires frequent recharging or replacement, which is costly and time-consuming. Researchers have proposed energy harvesting systems that capture sustainable ambient energy from the environment to address this issue. This paper presents a hybrid system for harvesting sustainable energy from solar and wind sources. The system features a boost converter controlled by a novel hybrid method combining the Honey Badger Algorithm (HBA) and Harris Hawks Optimization (HHO). This method maximizes power extraction from solar and wind sources, enhancing overall system efficiency. Additionally, the system includes an innovative energy management algorithm that selects the most powerful input source while protecting the storage battery from overcharging or complete depletion, thereby extending its lifespan. The proposed design is validated through MATLAB/Simulink simulations. The HHO-HBA MPPT is compared with existing MPPT methods, evaluating efficiency, battery charge curves, and IoT network energy status. Simulation results show that the proposed approach significantly increases network longevity, offering a cost-effective and sustainable solution for the energy needs of Wireless Sensor Network (WSN)-IoT devices.

In coastal areas, saltwater and freshwater converge, creating high biodiversity and significant pollution issues from industrial waste, microplastics, fertilizers, and medications. These substances disrupt ecosystems and species' life cycles that regulate key physiological processes, even affecting humans. Chromatography is the most commonly used technique for detecting drugs in water, but it is also expensive and requires specialized personnel. This study used optical sensors around a tube to detect drugs (Dacortin, Diazepam, and Prednisone) and baking flour in water by measuring light absorption at different wavelengths (visible, infrared, and ultraviolet). Data changes were processed using an Arduino Uno microprocessor, configured as an IoT device. The concentrations were 0, 10, 15, and 30 mg/100 mL. The results display several models using MATLAB. K-nearest neighbor showed the best performance with 100% accuracy and the shortest training time (0.71 seconds). Ultraviolet sensors provided the most variable and impactful data, while red and green sensors showed similar results, and blue had higher dispersion. Infrared sensors were less sensitive, requiring future adjustment for improved analysis. Regression models showed high performance, the best being Rational Quadratic Gaussian Process Regression, achieving a Root Mean Squared Error of 0.2 mg/100 mL in training and 0.01 mg/100 mL in testing. Some training errors were attributed to outliers caused by electronic or sample issues, highlighting the need for outlier filtering. Test dataset errors were minimal and consistent with training results, validating the model's accuracy.

Essential oils (EOs) are chemically complex natural matrices whose quality and bioactivity are governed by structurally diverse secondary metabolites. Conventional techniques such as gas chromatography–mass spectrometry (GC–MS) provide detailed compositional profiling, yet they remain costly, labor-intensive, and unsuitable for real-time monitoring. Here, we present a multimodal chemometric framework that integrates UV–Vis–NIR spectroscopy (190–1100 nm) with low-cost metal oxide gas sensors to quantitatively predict EO metabolite concentrations using machine learning. Multivariate analyses employing t-distributed stochastic neighbor embedding (t-SNE), uniform manifold approximation and projection (UMAP), and correlation mapping revealed chemically coherent clustering of Cistus ladanifer EO samples and biochemical associations between sensor responses and metabolite families, highlighting the richness of the fused feature space. Metabolites identified by GC–MS were grouped into seven functional classes including terpenic hydrocarbons, sesquiterpenic hydrocarbons, alcohols, aldehydes, ketones, esters, and residuals. These groupings guided targeted regression modeling. Data validation was performed against GC–MS quantified metabolites. Among the tested algorithms, Ridge regression achieved the highest predictive performance (R2 = 0.999). Lasso regression followed with R2 = 0.971, favoring sparsity at the expense of completeness. Partial least squares algorithm failed to capture variance in the high-dimensional multimodal dataset. Feature attribution based on Shapley values demonstrated that accurate predictions required the joint contribution of distributed spectral bands and complementary sensor responses, underscoring the necessity of multimodal fusion for resolving chemically heterogeneous and low-abundance metabolites. This work establishes a scalable, non-destructive, and real-time strategy for EO profiling, with broad implications for traceability, sustainable cultivation, and smart agriculture, and illustrates the transformative role of machine learning in chemometric exploration of natural products.

Wireless multimedia sensor networks (WMSN) have been mainly applied in urban areas using Zigbee, Bluetooth or Wi-Fi to forward the data, but no application has been designed in rural areas due to the large distances and energy restrictions. The related literature shows only one deployment with LoRa using a Raspberry Pi 3 node, but no measure of real video delivery was performed. In this paper, we propose the use of LoRa technology for video streaming in a WMSN for smart agriculture. The objective of this design is to monitor the fields by having nodes with sensors, nodes with cameras, and aggregator nodes. Two nodes are used to test the performance of the network and ensure that video streaming with LoRa is possible. The emitter node, an Arduino ATMega, sends a video to the receiver node, an ESP32. The LoRa UART SX1278 was used in both nodes to communicate them. Results in the laboratory, with both nodes close, indicate that the minimum latency is achieved using a block size of 6408 bytes. Four video resolutions and 4 video extensions are tested to evaluate the maximum video quality which can be streamed. The results in a real scenario with nodes spaced 731 meters indicated that video in .mp4 video format at 240 p resolution can be streamed correctly with a medium access time of 3.2 s and 0% of packet loss for a video of 5 s. Thus, it is possible to affirm that LoRa can be used for video streaming.

Monitoring ocean salinity and temperature is vital for understanding ecosystem health and mitigating marine pollution from untreated municipal sewage. This paper explores the use of IoT systems for continuous underwater monitoring of marine outfalls from wastewater treatment plants, providing early warnings of contamination. The system is composed of multiple sensors, highlighting the temperature and salinity sensors in this paper. We have used a commercial temperature sensor, the DS18B20 and a tailored salinity sensor. Those sensors are connected to the sensor node, an ESP32. The node collects the data and sends it to a database. The temperature and salinity sensors are calibrated and validated with generated samples. The samples have a variation of salinity from 0 to 40 g/L, and the variation in temperature goes from 40 to 10 degrees C. The results indicated that the commercial temperature sensor has a bias, and we proposed an equation to correct it. The salinity sensor is calibrated, obtaining an equation with an R2 of 99.5 %. The network traffic analysis indicates an increase in traffic when errors in TPC packets appear.

Soil degradation issues, such as salinization, waterlogging, and erosion, pose significant challenges in agricultural irrigation and fertigation. Conventional laboratory methods for analyzing soil conditions are often cost-prohibitive, time-intensive, and require specialized personnel. This study investigates the application of a virtual sensor, incorporating RGB and LDR modules, to measure fertilizer concentration and water content in soil, as well as to differentiate between water and fertilizer. Statistical analysis of the acquired data using ANOVA revealed significant differences between samples under blue, yellow, and magenta lights, with p-values of 0.0425, 0.018, and 0.0425, respectively. Machine Learning (ML) techniques were employed to enhance the system's performance, yielding an 87.5% accuracy in binary classification using the K-Nearest Neighbors (KNN) and Euclidean models. Additionally, regression models demonstrated R2 values of 0.802 and 0.891 for irrigated and ferti-irrigated soils, respectively. These results suggest that the proposed optical sensor system is a promising, cost-effective alternative to traditional methods, though further validation in real-world applications is necessary.

Turbidity is one of the crucial parameters of water quality. Even though many commercial devices, low-cost sensors, and remote sensing data can efficiently quantify turbidity, they are not valid tools for the classification it. In this paper, we design, calibrate, and test a novel optical low-cost sensor for turbidity quantification and classification. The sensor is based on an RGB light source and a light detector. The analyzed samples are characterized by turbidity values from 0.02 to 60 NTUs, and have four different sources. These samples were generated to represent natural turbidity sources and leaves in the marine areas close to agricultural lands. The data are gathered using 64 different combinations of light, generating complex matrix data. Machine learning models are compared to analyze this data, including training, validation, and test datasets. Moreover, different alternatives for data preprocessing and feature selection are assessed. Concerning the quantification of turbidity, the best results were obtained using averaged data and principal components analyses in conjunction with exponential gaussian process regression, achieving an R2 of 0.979. Regarding the classification of the turbidity, an accuracy of 91.23% is obtained with the fine K-Nearest-Neighbor classifier. The cases in which data were misclassified are characterized by turbidity values lower than 5 NTUs. The obtained results represent an improvement over the current solutions in terms of turbidity quantification and a completely novel approach to turbidity classification.