Greenhouses can grow many off‐season vegetables and fruits, which improves people's quality of life. Greenhouses can also help crops resist natural disasters and ensure the stable growth of crops. However, it is highly challenging to carefully control the greenhouse climate. Therefore, the proposal of a greenhouse climate prediction model provides a way to solve this challenge. We focus on the six climatic factors that affect crops growth, including temperature, humidity, illumination, carbon dioxide concentration, soil temperature and soil humidity, and propose a GCP_lstm model for greenhouse climate prediction. The climate change in greenhouse is nonlinear, so we use long short‐term memory (LSTM) model to capture the dependence between historical climate data. Moreover, the short‐term climate has a greater impact on the future trend of greenhouse climate change. Therefore, we added a 5‐min time sliding window through the analysis experiment. In addition, sensors sometimes collect wrong climate data. Based on the existence of abnormal data, our model still has good robustness. We experienced our method on the data sets of three vegetables: tomato, cucumber and pepper. The comparison shows that our method is better than other comparison models.

A long short-term memory-based model for greenhouse climate prediction

The advancement of Internet of Things (IoT) technologies leads to a wide penetration and large-scale deployment of IoT systems across an entire city or even country. While IoT systems are capable of providing intelligent services, the large amount of data collected and processed in IoT systems also raises serious security concerns. Many research efforts have been devoted to design intelligent network intrusion detection system (NIDS) to prevent misuse of IoT data across smart applications. However, existing approaches may suffer from the issue of limited and imbalanced attack data when training the detection model, which make the system vulnerable especially for those unknown type attacks. In this study, a novel hierarchical adversarial attack (HAA) generation method is introduced to realize the level-aware black-box adversarial attack strategy, targeting the graph neural network (GNN)-based intrusion detection in IoT systems with a limited budget. By constructing a shadow GNN model, an intelligent mechanism based on a saliency map technique is designed to generate adversarial examples by effectively identifying and modifying the critical feature elements with minimal perturbations. A hierarchical node selection algorithm based on random walk with restart (RWR) is developed to select a set of more vulnerable nodes with high attack priority, considering their structural features, and overall loss changes within the targeted IoT network. The proposed HAA generation method is evaluated using the open-source data set UNSW-SOSR2019 with three baseline methods. Comparison results demonstrate its ability in degrading the classification precision by more than 30% in the two state-of-the-art GNN models, GCN and JK-Net, respectively, for NIDS in IoT environments. 

Hierarchical Adversarial Attacks Against Graph-Neural-Network-Based IoT Network Intrusion Detection System

Surveillance systems, in spite of the recent advances, still poses many challenges, especially in the field of patrolling or tracking of subjects through CCTV footage or any other portable drone mechanisms. Real time monitoring of public places for possible suspects are still made through manual observations in many places. The manual labour involved and the human errors that can occur makes the system less efficient. Many research attempts were made for designing foolproof systems and still going on, understanding the importance of the problem. This paper presents the implementation of an intelligent multicamera Face Recognition based surveillance system using FaceNet and MTCNN algorithm on Jetson TX2. The proposed portable system tracks the subject or the suspect with the camera ID/location together with the timestamp and logs his presence in the database, using multiple camera installation.

Face Recognition based Surveillance System Using FaceNet and MTCNN on Jetson TX2

Trust management-based and energy efficient hierarchical routing protocol in wireless sensor networks

Graph Neural Networks (GNNs) have emerged as a powerful and flexible framework for representation learning on irregular data. As they generalize the operations of classical CNNs on grids to arbitrary topologies, GNNs also bring much of the implementation challenges of their Euclidean counterparts. Model size, memory footprint, and energy consumption are common concerns for many real-world applications. Network binarization allocates a single bit to parameters and activations, thus dramatically reducing the memory requirements (up to 32x compared to single-precision floating-point numbers) and maximizing the benefits of fast SIMD instructions on modern hardware for measurable speedups. However, in spite of the large body of work on binarization for classical CNNs, this area remains largely unexplored in geometric deep learning. In this paper, we present and evaluate different strategies for the binarization of graph neural networks. We show that through careful design of the models, and control of the training process, binary graph neural networks can be trained at only a moderate cost in accuracy on challenging benchmarks. In particular, we present the first dynamic graph neural network in Hamming space, able to leverage efficient k-NN search on binary vectors to speed-up the construction of the dynamic graph. We further verify that the binary models offer significant savings on embedded devices. Our code is publicly available on Github1.

/pdf/binary-graph-neural-networks-h1f45f6kew.pdf

Binary Graph Neural Networks

Traditional neural networks usually concentrate on temporal data in system simulation, and lack of capabilities to reason inner logic relations between different dimensions of data collected from embedded sensors. This paper proposes a graph neural network-based modeling approach for IoT equipment (called GNNM-IoT), which considers both temporal and inner logic relations of data, in which vertices denote sensor data and edges denote relationships between vertices. The GNNM-IoT model's relationships between sensors with neural networks to produce nonlinear complex relationships. We have evaluated the GNNM-IoT using air-conditioner data from a world leading IoT company, which demonstrates that it is effective and outperforms ARIMA and LSTM methods.

/pdf/modeling-iot-equipment-with-graph-neural-networks-2lsp31ta8x.pdf

Modeling IoT Equipment With Graph Neural Networks

Real-time intelligent video processing on embedded devices with low power consumption can be useful for applications like drone surveillance, smart cars, and more. However, the limited resources of embedded devices is a challenging issue for effective embedded computing. Most of the existing work on this topic focuses on single device based solutions, without the use of cloud computing mechanisms for parallel processing to boost performance. In this paper, we propose a cloud platform for real-time video processing based on embedded devices. Eight NVIDIA Jetson TX1 and three Jetson TX2 GPUs are used to construct a streaming embedded cloud platform (SECP), on which Apache Storm is deployed as the cloud computing environment for deep learning algorithms (Convolutional Neural Networks - CNNs) to process video streams. Additionally, self-managing services are designed to ensure that this platform can run smoothly and stably, in the form of a metric sensor, a bottleneck detector and a scheduler. This platform is evaluated in terms of processing speed, power consumption, and network throughput by running various deep learning algorithms for object detection. The results show the proposed platform can run deep learning algorithms on embedded devices while meeting the high scalability and fault tolerance required for real-time video processing.

/pdf/a-streaming-cloud-platform-for-real-time-video-processing-on-30ox0dyibi.pdf

A Streaming Cloud Platform for Real-Time Video Processing on Embedded Devices

It is important for monitoring and predicting equipment failures. The existing fault prediction method has poor efficiency and accuracy on processing imbalanced data. This paper proposes a feature pattern-based LSTM method (called FLSTM, Feature based Long Short Term Memory) to analyze failures through processing imbalanced data. The method constructs a time-series feature matrix as the input to the LSTM model. In addition, we propose a failure prediction system based on Hadoop environment. The experimental results show that the FLSTM can improve failure prediction with imbalanced big data and the failure prediction system performs well.

FLSTM: Feature Pattern-Based LSTM for Imbalanced Big Data Analysis

Traditional character segmentation has low accuracy for container scene text recognition. Convolutional recurrent neural network (CRNN) and connectionist text proposal network (CTPN) methods cannot extract container text features effectively. This paper proposes a novel Container Text Detection and Recognition Network (CTDRNet) for accurately detecting and recognizing container scene text. The CTDRNet consists of three components: (1) CTDRNet text detection enables to improve detection accuracy for single words; (2) CTDRNet text recognition has faster convergence speed and detection accuracy; (3) CTDRNet post-processing improves detection and recognition accuracy. In the end, the CTDRNet is implemented and evaluated with an accuracy of 96% and processing rate of 2.5 fps.

/pdf/deep-learning-based-container-text-recognition-1lfrq334tt.pdf

Deep Learning Based Container Text Recognition

Multi-source heterogeneous data are playing increasingly important roles. It becomes important to harness multi-source heterogeneous data for effectively managing and mining data intelligence. This paper presents a hybrid data fusion model called FD-DFM (Feature and Decision-Data Fusion Model) based on deep learning. The FD-DFM integrates feature fusion and decision fusion into neural networks with the D-S evidence theory. The experiment shows that the FD-DFM model has higher accuracy than other existing methods with fruit recognition.

Liang Xu

Papers

Modeling IoT Equipment With Graph Neural Networks

A Streaming Cloud Platform for Real-Time Video Processing on Embedded Devices

FLSTM: Feature Pattern-Based LSTM for Imbalanced Big Data Analysis

Deep Learning Based Container Text Recognition

A Deep Learning-Based Hybrid Data Fusion Method for Object Recognition