Benameur Rabaie, Dahane Amine, Kechar Bouabdellah, Benyamina Abou El Hassan
Research Laboratory in Industrial Computing and Networks (RIIR), University of Oran 1, B.P. 1524, El M'Naouer, Oran 31000, Algeria.
Institute of Applied Science and Technology, ISTA, University of Oran 1, B.P. 1524, El M'Naouer, Oran 31000, Algeria.
Sensors (Basel). 2024 Feb 10;24(4):1162. doi: 10.3390/s24041162.
The agricultural sector faces several difficulties today in ensuring the safety of food supply, including water scarcity. This study presents the design and development of a low-cost and full-featured fog-IoT/AI system targeted towards smallholder farmer communities (SFCs). However, the smallholder community is hesitant to adopt technology-based solutions. There are many overwhelming reasons for this, but the high cost, implementation complexity, and malfunctioning sensors cause inappropriate decisions. The PRIMA INTEL-IRRIS project aims to make digital and innovative agricultural technologies more appealing and available to these communities by advancing the intelligent irrigation "in-the-box" concept. Considered a vital resource, collected data are used to detect anomalies or abnormal behavior, providing information about an occurrence or a node failure. To prevent agro-field data leakage, this paper presents an innovative, smart, and sustainable low-cost irrigation system that employs artificial intelligence (AI) techniques to analyze anomalies and problems in water usage. The sensor anomaly can be detected using an autoencoder (AE) and a generative adversarial network (GAN). We will feed the autoencoders' anomaly detection models with time series records from the datasets and replace detected anomalies with the reconstructed outputs. When integrated with an IoT platform, this methodology is a tool for easing the labeling of sensor anomalies and can help create supervised datasets for future research. In addition, anomalies can be corrected by prediction models based on deep learning approaches, applying CNN/BiLSTM architecture. The results show that AEs outperform the GANs, achieving an accuracy of 90%, 95%, and 97% for soil moisture, air temperature, and air humidity, respectively. The proposed system is designed to ensure that the data are of high quality and reliable enough to make sound decisions compared to the existing platforms.
如今，农业部门在确保粮食供应安全方面面临诸多困难，其中包括水资源短缺。本研究介绍了一种针对小农户社区（SFCs）设计和开发的低成本、全功能的雾物联网/人工智能系统。然而，小农户社区对采用基于技术的解决方案持犹豫态度。造成这种情况的原因有很多，但成本高昂、实施复杂以及传感器故障会导致决策不当。PRIMA INTEL - IRRIS项目旨在通过推进智能灌溉“即装即用”概念，使数字和创新农业技术对这些社区更具吸引力且更易获取。收集到的数据被视为一项重要资源，用于检测异常或异常行为，提供有关事件或节点故障的信息。为防止农田数据泄露，本文提出了一种创新、智能且可持续的低成本灌溉系统，该系统采用人工智能（AI）技术来分析用水中的异常和问题。可以使用自动编码器（AE）和生成对抗网络（GAN）来检测传感器异常。我们将把数据集的时间序列记录输入自动编码器的异常检测模型，并将检测到的异常替换为重建输出。当与物联网平台集成时，这种方法是一种便于标记传感器异常的工具，有助于创建用于未来研究的监督数据集。此外，可以通过基于深度学习方法的预测模型（应用CNN / BiLSTM架构）来纠正异常。结果表明，自动编码器在检测土壤湿度、气温和空气湿度异常方面的表现优于生成对抗网络，准确率分别达到90%、95%和97%。与现有平台相比，所提出的系统旨在确保数据具有高质量且足够可靠，以便做出合理决策。
