Input Channels Research Articles

Energy, economic and environmental feasibility of energy recovery from wastewater treatment plants in mountainous areas: A case study of gharyan city – LIBYA

Wastewater treatment facilities at high places can give chances for renewable and sustainable energy generation by putting hydroelectric turbines at the input and drain channels of wastewater treatment plants, and they can also use the sludge generated during the treatment process to make biogas, which can be used to generate power. Purified water is subsequently used to irrigate decorative plants along highways, in gardens, and in woods. The fermentation wastes are utilized as organic fertilizer to improve agricultural soil quality. At the Gharyan sewage station, a hybrid system consisting of a hydroelectric station and an electric generator powered by biogas is proposed in this research. This is because the city is distinguished by its high location, which is approximately 713 m above sea level. The results showed that the proposed system would provide an electric power of 490 kW, which is sufficient to cover 87.5% of the electrical energy consumption of the station. The amount of treated water is approximately 13,000 m3/day, and the amount of organic fertilizer is about 17 tons/day. The investment value is anticipated to be around $1,478,000, while the cost of producing a unit of electric energy is expected to be 2.83 ¢/kWh. This system's yearly net profit is predicted to be $307,765. The capital's recovery period is anticipated to be 3.44 years. The planned hybrid system will limit the discharge of an estimated 1,886 tons of CO2 gas each year.

A new identification method of economic large shocks in the input–output framework: application to the COVID-19

The COVID-19 pandemic has had a significant and unprecedented impact on the global economy, affecting both supply and demand in various industries. The literature on the pandemic has mainly focused on evaluating its real-time impact, using input–output analysis as a toolset. However, this paper aims to identify the magnitude of the large shock after observing its impacts on the economy. The proposed methodology adopts the main assumptions usually made in input–output analysis and focuses on the short term, assuming no significant changes in structures and prices. The approach is based on the technique of hypothetical extraction, which is generalized to consider the possibility of a shock simultaneously impacting all industries. Applying this approach to the Moroccan economy to identify the COVID-19 shock at the sectoral level, it appears that the most significant negative shocks were experienced by three industries, namely tourism, transport and mechanical industries, metallurgical, electrical. This is largely due to the nature of these industries, which are more sensitive to production constraints imposed to limit the spread of the pandemic. The results also show that other branches of activity have been severely impacted via the indirect channel of inputs rather than directly by the shock of the pandemic.

Data-Driven Control for Discrete-Time Nonlinear Systems With Dual-Channel Dynamic Event-Triggered Mechanism

The problem of data-driven control (DDC) for nonlinear systems with dual-channel event-triggering is investigated in this article. First, the nonlinear system is converted into a dynamic linearized data model using the local dynamic linearization techniques. Then, a new dual-channel triggered control framework is designed based on the transformed linearized data model, and semi-independent dynamic event-triggered strategies are designed for the two channels of control input and measurement output. Different from the traditional single-channel event triggered, the main advantages of the proposed control strategy are that the input and output channels are semi-independently triggered, and it can accomplish the tracking control objectives by relying only on the input and output data of the system, and reducing the communication frequency of the input and output channels while ensuring stability. Finally, the effectiveness of the developed dual-channel event-triggered control strategy is verified by the application experiments.

BLIND IDENTIFIABILITY OF A LINEAR DYNAMIC COMMUNICATION CHANNEL MODEL

The article describes the problem of developing a mathematical model of data transmission channel in communication systems in the case when signals at the channel input channel are unknown at the channel output. Development of mathematical models directly from observed data constitutes an identification problem, and the case under consideration belongs to the blind identification problem. The work consistently examines the concept of linear dynamic system identifiability in state space and shows the connection of this concept with the observability and controllability concepts. A number of statements about identifiability of a linear dynamic system describing communication channel are proved. In particular, it is shown that if the system is identifiable, it is sufficient to use the input-output model for its description, while the input-state-output model is redundant. Further, based on the proven statements, blind identifiability conditions are formulated for the case of a discrete system with one input and multiple outputs, which are imposed on the components of the system’s transfer matrix.

Sensorless Saliency-Based Control of Dual Induction Machines Under Dynamic Load Imbalances Using Three Current Sensors

To perform vector control of single-inverter dual induction motors, drives are standardly equipped with current and position sensors. In medium-power railway applications, a shaft encoder is usually mounted on each motor, and the inverter output current is measured using two current sensors. Yet, this configuration features some disadvantages. First, torque-sharing cannot be computed because stator current of each motor is unknown. Similarly, saliency extraction of each motor is not achievable since individual current drawn by each motor is not measured. An option that achieves torque-sharing calculation is to attach two current sensors to each motor, provided that both rotor positions are known. However, industrial inverter/control platforms considered offer a maximum of three input channels for current sensors. Besides, some industry fields tend towards encoderless speed control. Therefore, this paper proposes to use three current sensors in total and no shaft encoders. Thanks to a special current sensor arrangement and a voltage step excitation method, saliency-based encoderless control is achieved. In addition, a method is presented to estimate torque-sharing. Experimental measurements, taken on a dual induction motor test bench, will show the capability of the proposed methodology to identify individual loading as well as excellent encoderless control performance.

Changes of sediment provenance driven by the sea level and Indian summer monsoon in the northern Ninetyeast Ridge over the past 50 kyr

The Ninetyeast Ridge is situated within the world's largest sediment “source-sink” system, which extends from the Tibetan Plateau to the Bay of Bengal (BoB). Modest and stable sedimentation rate makes it a suitable region for studying long-term land-sea interaction. Recent research has primarily focused on the northern Bengal Fan, leaving limited knowledge about sedimentary processes on the distant Ninetyeast Ridge. This study aims to address this gap by investigating sediment transport pattern and its response to climatic and environmental changes since the last glacial period, using the gravity core (CJ04–50) selected from the northern Ninetyeast Ridge. The provenance discrimination based on δEu-(Gd/Yb)N revealed that the Irrawaddy (Irr) River, along with the Ganges-Brahmaputra (G-B) Rivers, represented the primary sources of the study area. Quantitative analysis equations indicated that the contribution of the G-B Rivers decreased from the last glacial period (22%) to the Holocene period (16%), while that of the Irr River increased (from 78% to 84%). Chemical index of alteration (CIA*) values, and terrigenous input proxy indicators (Ti/Ca, Rb/Sr) were used to indicate the changes of weathering and terrigenous input over the past 50 kyr. During the last glacial period, CIA*, Ti/Ca and Rb/Sr revealed significantly millennial-scale periodic changes, including the distinct declining trend during the Last Glacial Maximum (LGM) and Heinrich 2–4 (H2-H4) periods, followed by recovery trend during subsequent warm periods. During these cold periods of LGM and H2–H4, weakened Indian monsoon condition, reduced precipitation, cold climate, and expanding ice sheets, adversely affected physical erosion and chemical weathering in the source area, and the reduced river runoff impeded sediment transport. Conversely, these environmental factors reversed during relatively warm periods. Since the Holocene, CIA*, Ti/Ca and Rb/Sr declined, which was contrast to the climate change, and the sea level changes became the dominant factor. The rapidly rising sea level submerged the main terrigenous input channel (the Swatch of No Ground) of the G-B Rivers, made the depositional center shift from the Bengal Fan to the continental shelf, causing the decline of terrigenous input and lower CIA* values.

Cross channel weight sharing for image classification

Hypercomplex convolutional neural networks (HCNNs) have recently been used to improve deep learning architectures due to their ability to share weights across input channels and thus improve the cohesiveness of learned representations within the layers. Much work has been done to study the effect of hypercomplex representation in CNNs. However, to date, none of these models have used fully hypercomplex architectures, meaning that some of the layers in these earlier networks used conventional real-valued calculations that did not engage cross-channel weight sharing. Here, we introduce and study full hypercomplex CNNs by ensuring all layers perform hypercomplex calculations. For the earlier HCNNs, the real-valued layers are found in: (I) in the front end of the networks, (II) in the back end of the networks, and (III) in the residual blocks of the networks. The present research examines the performance of HCNNs when the modules mentioned above are replaced with hypercomplex equivalents. These representational networks have outperformed and shown state-of-the-art results compared to previous hypercomplex models (on some specific datasets). A disadvantage of these representational networks is that they consume high computational costs. To reduce this cost, novel separable hypercomplex networks (SHNNs) are proposed. They are created by factoring a quaternion convolutional module into two consecutive separable vectormap convolutional modules. As the successive layers of deep HCNNs perform local hierarchical grouping on increasingly abstract features, these groupings are responsible for the long-distance interaction problems found in the dense layers. To handle these problems, researchers have applied attention mechanisms in hypercomplex space.This paper offers a perspective on the basic concepts of representational networks and attention-based representational networks with the focus on (1) reviewing deep full HCNNs, including hypercomplex-based fully connected dense layer; (2) analyzing hypercomplex-based separable concept; (3) describing how the hypercomplex spatial layer of residual bottleneck block is replaced with an attention layer; and (4) providing a comprehensive summary comparing the performance of recent deep HCNNs, deep convolutional neural networks, and original attention based networks. We compare the performance of the networks in terms of accuracy and the number of trainable parameters on several image classification datasets.

Disturbance observer‐based PID synchronization control of complex networks with probabilistic interval time‐varying delays and multiple disturbances

AbstractIn this article, the disturbance observer‐based proportional–integral–derivative (PID) control protocol is investigated for synchronization of complex networks with probabilistic interval time‐varying delays and multiple disturbances. For the sake of reducing conservatism, a method of probabilistic interval time‐varying delay is proposed by introducing two certain intervals with known probability distributions. Moreover, two different disturbances are taken into account. One of the disturbances is produced by an exogenous system which acts through the input channel, while the other is usual norm‐bounded disturbance. Simultaneously, a disturbance observer is designed to estimate and compensate the disturbance generated by the exogenous system. By constructing an appropriate Lyapunov–Krasovskii function, a sufficient criterion is obtained to ensure both the exponential synchronization and prescribed performance index. Finally, the simulation examples are employed to demonstrate the validity of the theoretical results.

Basis scaling and double pruning for efficient inference in network-based transfer learning

Network-based transfer learning allows the reuse of deep learning features with limited data, but the resulting models can be unnecessarily large. Although network pruning can improve inference efficiency, existing algorithms usually require fine-tuning that may not be suitable for small datasets. In this paper, using the singular value decomposition, we decompose a convolutional layer into two layers: a convolutional layer with the orthonormal basis vectors as the filters, and a “BasisScalingConv” layer which is responsible for rescaling the features and transforming them back to the original space. As the filters in each decomposed layer are linearly independent, when using the proposed basis scaling factors with the Taylor approximation of importance, pruning can be more effective and fine-tuning individual weights is unnecessary. Furthermore, as the numbers of input and output channels of the original convolutional layer remain unchanged after basis pruning, it is applicable to virtually all architectures and can be combined with existing pruning algorithms for double pruning to further increase the pruning capability. When transferring knowledge from ImageNet pre-trained models to different target domains, with less than 1% reduction in classification accuracies, we can achieve pruning ratios up to 74.6% for CIFAR-10 and 98.9% for MNIST in model parameters.

Deep learning-based outcome prediction using PET/CT and automatically predicted probability maps of primary tumor in patients with oropharyngeal cancer

Background and ObjectiveRecently, deep learning (DL) algorithms showed to be promising in predicting outcomes such as distant metastasis-free survival (DMFS) and overall survival (OS) using pre-treatment imaging in head and neck cancer. Gross Tumor Volume of the primary tumor (GTVp) segmentation is used as an additional channel in the input to DL algorithms to improve model performance. However, the binary segmentation mask of the GTVp directs the focus of the network to the defined tumor region only and uniformly. DL models trained for tumor segmentation have also been used to generate predicted tumor probability maps (TPM) where each pixel value corresponds to the degree of certainty of that pixel to be classified as tumor. The aim of this study was to explore the effect of using TPM as an extra input channel of CT- and PET-based DL prediction models for oropharyngeal cancer (OPC) patients in terms of local control (LC), regional control (RC), DMFS and OS. MethodsWe included 399 OPC patients from our institute that were treated with definitive (chemo)radiation. For each patient, CT and PET scans and GTVp contours, used for radiotherapy treatment planning, were collected. We first trained a previously developed 2.5D DL framework for tumor probability prediction by 5-fold cross validation using 131 patients. Then, a 3D ResNet18 was trained for outcome prediction using the 3D TPM as one of the possible inputs. The endpoints were LC, RC, DMFS, and OS. We performed 3-fold cross validation on 168 patients for each endpoint using different combinations of image modalities as input. The final prediction in the test set (100) was obtained by averaging the predictions of the 3-fold models. The C-index was used to evaluate the discriminative performance of the models. ResultsThe models trained replacing the GTVp contours with the TPM achieved the highest C-indexes for LC (0.74) and RC (0.60) prediction. For OS, using the TPM or the GTVp as additional image modality resulted in comparable C-indexes (0.72 and 0.74). ConclusionsAdding predicted TPMs instead of GTVp contours as an additional input channel for DL-based outcome prediction models improved model performance for LC and RC.

Modelling and numerical analysis of combined dual impact of SRS and FWM on the performance of upstream and downstream channels of a novel UDWDM/DWDM-PON

Wavelength division multiplexing (WDM) is an important method in modern optical communication systems for both long-haul and access networks. Particularly, after being standardized by Telecommunication Standardization Sector of International Telecommunication Union (ITU-T), WDM passive optical network (WDM-PON) becomes a considerable choice for fronthaul implementations of 5G networks. Stimulated Raman scattering (SRS) and four-wave mixing (FWM) are crucial nonlinear impacts that significantly limit the performance of WDM-based optical fiber communication systems. In this paper, an ultra-dense/dense WDM-PON (UDWDM/DWDM-PON) architecture has been proposed and combined dual impact of SRS and FWM on performances of upstream channels (USCs) and downstream channels (DSCs) of the proposed UDWDM/DWDM-PON has been numerically analyzed. USCs have been assumed to operate in 1310 nm region while operating wavelength region of DSCs has been taken as 1550 nm. Simulations have been performed on both UDWDM-PON architectures with 3.125 GHz and 6.25 GHz channel spacings and DWDM-PON architectures with channel spacings between 12.5 GHz and 100 GHz. 2x15-channel and 2x63-channel system structures have been considered in simulations. Signal-to-crosstalk ratio (SXR) has been chosen as performance parameter and its variation with channel input powers, channel spacings and channel lengths have been simulated for both USCs and DSCs of 2x15- and 2x63-channel UDWDM/DWDM-PONs. Results show that the proposed UDWDM/DWDM-PONs using single G.652 fiber can exhibit a reliable communication with minimum 23 dB SXR performance under combined dual impact of SRS and FWM unless some highlights about input powers and channel lengths mentioned in the paper are neglected.

Automatic shoreline detection by processing planview timex images using bi-LSTM networks

A new automatic shoreline detection method by using a bidirectional Long Short-Term Memory (bi-LSTM) Network that processes images column by column is presented. The model is trained on manually extracted shorelines from time-exposure video-images and is very robust against the selection of images for training. Thanks to the novelty of working with image columns, instead of with the whole image, the amount of labelled images for training is limited to a few tens or even less if the conditions are good. Moreover, this column approach makes the model to be robust to variable illuminated images and more easily interpretable, light and fast. There is a wide range of configuration parameters for the bi-LSTM layer by which the system works correctly, which facilitate to use the same network in different video stations. The highest accuracy is obtained by using CIELAB colour space. Without pre-processing the raw colour channels or defining a region of interest and without post-processing the obtained shorelines, the model demonstrates impressive accuracy with mean errors of 2.8 pixels (1.4 m) in Castelldefels and 1.7 pixels (0.85 m) in Barcelona. The method could also be effective for satellite shoreline detection by using as input channel the water index of the satellite detection techniques.

FD-CNN: A Frequency-Domain FPGA Acceleration Scheme for CNN-Based Image-Processing Applications

In the emerging edge-computing scenarios, FPGAs have been widely adopted to accelerate convolutional neural network (CNN)–based image-processing applications, such as image classification, object detection, and image segmentation, and so on. A standard image-processing pipeline first decodes the collected compressed images from Internet of Things (IoTs) to RGB data, then feeds them into CNN engines to compute the results. Previous works mainly focus on optimizing the CNN inference parts. However, we notice that on the popular ZYNQ FPGA platforms, image decoding can also become the bottleneck due to the poor performance of embedded ARM CPUs. Even with a hardware accelerator, the decoding operations still incur considerable latency. Moreover, conventional RGB-based CNNs have too few input channels at the first layer, which can hardly utilize the high parallelism of CNN engines and greatly slows down the network inference. To overcome these problems, in this article, we propose FD-CNN, a novel CNN accelerator leveraging the partial-decoding technique to accelerate CNNs directly in the frequency domain. Specifically, we omit the most time-consuming IDCT (Inverse Discrete Cosine Transform) operations of image decoding and directly feed the DCT coefficients (i.e., the frequency data) into CNNs. By this means, the image decoder can be greatly simplified. Moreover, compared to the RGB data, frequency data has a narrower input resolution but has 64× more channels. Such an input shape is more hardware friendly than RGB data and can substantially reduce the CNN inference time. We then systematically discuss the algorithm, architecture, and command set design of FD-CNN. To deal with the irregularity of different CNN applications, we propose an image-decoding-aware design-space exploration (DSE) workflow to optimize the pipeline. We further propose an early stopping strategy to tackle the time-consuming progressive JPEG decoding. Comprehensive experiments demonstrate that FD-CNN achieves, on average, 3.24×, 4.29× throughput improvement, 2.55×, 2.54× energy reduction and 2.38×, 2.58× lower latency on ZC-706 and ZCU-102 platforms, respectively, compared to the baseline image-processing pipelines.

Steady-state forms of channel profiles shaped by debris flow and fluvial processes

Abstract. Debris flows regularly traverse bedrock channels that dissect steep landscapes, but our understanding of bedrock erosion by debris flows and their impact on steepland morphology is still rudimentary. Quantitative models of steep bedrock channel networks are based on geomorphic transport laws designed to represent erosion by water-dominated flows. To quantify the impact of debris flow erosion on steep channel network form, it is first necessary to develop methods to estimate spatial variations in bulk debris flow properties (e.g., flow depth, velocity) throughout the channel network that can be integrated into landscape evolution models. Here, we propose and evaluate two methods to estimate spatial variations in bulk debris flow properties along the length of a channel profile. We incorporate both methods into a model designed to simulate the evolution of longitudinal channel profiles that evolve in response to debris flow and fluvial processes. To explore this model framework, we propose a general family of debris flow erosion laws where erosion rate is a function of debris flow depth and channel slope. Model results indicate that erosion by debris flows can explain the occurrence of a scaling break in the slope–area curve at low-drainage areas and that upper-network channel morphology may be useful for inferring catchment-averaged erosion rates in quasi-steady landscapes. Validating specific forms of a debris flow incision law, however, would require more detailed model–data comparisons in specific landscapes where input parameters and channel morphometry can be better constrained. Results improve our ability to interpret topographic signals within steep channel networks and identify observational targets critical for constraining a debris flow incision law.

Performance Analysis of Chicken Freshness classification using Naïve Bayes, Decision Tree, and k-NN

Chicken is one of the staple foods that is widely enjoyed by all. To obtain the benefits of chicken meat, the level of freshness becomes one of the main keys. In general, the level of freshness of chicken meat is divided into two classes, namely fresh and non-fresh. The difference in the level of freshness can be seen from the color changes of each class. Spoiled chicken (chicken died yesterday) is one type of meat in the non-fresh group. The widespread sale of spoiled chicken meat among the public raises doubts about choosing chicken that is suitable and unsuitable for consumption. Therefore, chicken meat freshness classification is needed to facilitate the selection of chicken meat based on color characteristics. The use of Naive Bayes Classifier algorithm in categorizing fresh and non-fresh classes is done by calculating the probability value of each image channel input. This research was conducted to compare the Naive Bayes, decision tree, and K-NN algorithms in classifying chicken meat based on color characteristics. The results of the study showed that the Naive Bayes classifier algorithm was superior to the decision tree and K-NN algorithms with an accuracy rate of 75%, precision of 79%, and recall of 65%. It is known that 27 images were predicted correctly and 9 images were predicted incorrectly out of a total 36 data. The use of a histogram in this study aims to differentiate chicken meat images from non-meat during the testing process of the model using the Naive Bayes classifier algorithm.

Odor representations from the two nostrils are temporally segregated in human piriform cortex

The human olfactory system has two discrete channels of sensory input, arising from olfactory epithelia housed in the left and right nostrils. Here, we asked whether the primary olfactory cortex (piriform cortex [PC]) encodes odor information arising from the two nostrils as integrated or distinct stimuli. We recorded intracranial electroencephalogram (iEEG) signals directly from PC while human subjects participated in an odor identification task where odors were delivered to the left, right, or both nostrils. We analyzed the time course of odor identity coding using machine-learning approaches and found that uni-nostril odor inputs to the ipsilateral nostril are encoded ∼480-ms faster than odor inputs to the contralateral nostril on average. During naturalistic bi-nostril odor sampling, odor information emerged in two temporally segregated epochs, with the first epoch corresponding to the ipsilateral and the second epoch corresponding to the contralateral odor representations. These findings reveal that PC maintains distinct representations of odor input from each nostril through temporal segregation, highlighting an olfactory coding scheme at the cortical level that can parse odor information across nostrils within the course of a single inhalation.

Deep Learning Framework with Multi-Head Dilated Encoders for Enhanced Segmentation of Cervical Cancer on Multiparametric Magnetic Resonance Imaging.

T2-weighted magnetic resonance imaging (MRI) and diffusion-weighted imaging (DWI) are essential components of cervical cancer diagnosis. However, combining these channels for the training of deep learning models is challenging due to image misalignment. Here, we propose a novel multi-head framework that uses dilated convolutions and shared residual connections for the separate encoding of multiparametric MRI images. We employ a residual U-Net model as a baseline, and perform a series of architectural experiments to evaluate the tumor segmentation performance based on multiparametric input channels and different feature encoding configurations. All experiments were performed on a cohort of 207 patients with locally advanced cervical cancer. Our proposed multi-head model using separate dilated encoding for T2W MRI and combined b1000 DWI and apparent diffusion coefficient (ADC) maps achieved the best median Dice similarity coefficient (DSC) score, 0.823 (confidence interval (CI), 0.595-0.797), outperforming the conventional multi-channel model, DSC 0.788 (95% CI, 0.568-0.776), although the difference was not statistically significant (p > 0.05). We investigated channel sensitivity using 3D GRAD-CAM and channel dropout, and highlighted the critical importance of T2W and ADC channels for accurate tumor segmentation. However, our results showed that b1000 DWI had a minor impact on the overall segmentation performance. We demonstrated that the use of separate dilated feature extractors and independent contextual learning improved the model's ability to reduce the boundary effects and distortion of DWI, leading to improved segmentation performance. Our findings could have significant implications for the development of robust and generalizable models that can extend to other multi-modal segmentation applications.

Disturbance-observer-based discrete-time sliding mode predictive control for overhead cranes with matched and unmatched disturbances

In actual working environments, underactuated overhead cranes are susceptible to the adverse effects of the mechanical friction, uncertain parameters, and other disturbances. In particular, if the payload is disturbed (i.e. the unmatched disturbance), unpredictable swings would occur. Therefore, a discrete time sliding mode predictive control for overhead cranes based on a disturbance observer is proposed. Specifically, to eliminate uncertain disturbances, we propose a disturbance observer to estimate and compensate the disturbance to the input channel. Then, a new discrete sliding mode control law is obtained by adding the disturbance error estimator to the discrete sliding mode surface. In order to reduce the chattering inherent in the sliding mode, we integrate the discrete sliding mode law into the model predictive control, and use the rolling optimization and feedback correction of the predictive control, and finally output the optimal control law to act on the system. This control method can not only further improve the control accuracy, but also has sufficient robustness. Simulation and robustness experiments under various conditions verify the effectiveness of the proposed controller.

Lightbox: Sensor Attack Detection for Photoelectric Sensors via Spectrum Fingerprinting

Photoelectric sensors are utilized in a range of safety-critical applications, such as medical devices and autonomous vehicles. However, the public exposure of the input channel of a photoelectric sensor makes it vulnerable to malicious inputs. Several studies have suggested possible attacks on photoelectric sensors by injecting malicious signals. While a few defense techniques have been proposed against such attacks, they could be either bypassed or used for limited purposes. In this study, we propose Lightbox, a novel defense system to detect sensor attacks on photoelectric sensors based on signal fingerprinting. Lightbox uses the spectrum of the received light as a feature to distinguish the attacker’s malicious signals from the authentic signal, which is a signal from the sensor’s light source. We evaluated Lightbox against (1) a saturation attacker, (2) a simple spoofing attacker, and (3) a sophisticated attacker who is aware of Lightbox and can combine multiple light sources to mimic the authentic light source. Lightbox achieved the overall accuracy over 99% for the saturation attacker and simple spoofing attacker, and robustness against a sophisticated attacker. We also evaluated Lightbox considering various environments such as transmission medium, background noise, and input waveform. Finally, we demonstrate the practicality of Lightbox with experiments using a single-board computer after further reducing the training time.

Spatiotemporal Fusion Model of Remote Sensing Images Combining Single-Band and Multi-Band Prediction

In recent years, convolutional neural network (CNN)-based spatiotemporal fusion (STF) models for remote sensing images have made significant progress. However, existing STF models may suffer from two main drawbacks. Firstly, multi-band prediction often generates a hybrid feature representation that includes information from all bands. This blending of features can lead to the loss or blurring of high-frequency details, making it challenging to reconstruct multi-spectral remote sensing images with significant spectral differences between bands. Another challenge in many STF models is the limited preservation of spectral information during 2D convolution operations. Combining all input channels’ convolution results into a single-channel output feature map can lead to the degradation of spectral dimension information. To address these issues and to strike a balance between avoiding hybrid features and fully utilizing spectral information, we propose a remote sensing image STF model that combines single-band and multi-band prediction (SMSTFM). The SMSTFM initially performs single-band prediction, generating separate predicted images for each band, which are then stacked together to form a preliminary fused image. Subsequently, the multi-band prediction module leverages the spectral dimension information of the input images to further enhance the preliminary predictions. We employ the modern ConvNeXt convolutional module as the primary feature extraction component. During the multi-band prediction phase, we enhance the spatial and channel information captures by replacing the 2D convolutions within ConvNeXt with 3D convolutions. In the experimental section, we evaluate our proposed algorithm on two public datasets with 16x resolution differences and one dataset with a 3x resolution difference. The results demonstrate that our SMSTFM achieves state-of-the-art performance on these datasets and is proven effective and reasonable through ablation studies.