Cascade Neural Network Research Articles

Table tennis ball landing control in a robotic system by cameras

Abstract Controlling the landing position of a spinning ball is difficult when using a table tennis robot. A complete physical model requires the factoring in of aerodynamic elements and object collisions, and inaccurate environmental coefficients would increase the landing position error. This study proposed a landing position control method based on a cascade neural network (CNN) that consists of forward and recurrent neural networks (RNNs). The forward NNs are used to estimate the velocity of the outgoing ball according to the velocity and acceleration of the incoming ball captured by cameras and the desired velocity of the outgoing ball. The RNN is employed to reverse-predict ball displacement based on the state of the incoming ball, desired landing point, and ball flight duration. The experiments verified that the method proposed in this study achieved control of differently spinning balls more effectively than the locally weighted regression (LWR)-based model did. The success rate of the CNN at two of six desired landing points was 25.9% and 32.9% higher, respectively, compared with use of the LWR-based model.

Application of cascaded neural network for prediction of voltage stability margin in a solar and wind integrated power system

Voltage stability is a paramount concern in the management of renewable-rich power systems. As the diffusion of renewable energy sources continues to increase, accurately estimating voltage stability margins becomes essential. This paper addresses the critical need for effective voltage stability margin estimation and presents a novel approach to predict the same on the solar and wind integrated power system. A new machine learning methodology is developed based on the cascaded neural network (CNN). The proposed CNN methodology effectively estimates the loadability margin in a renewable-rich power system under normal operating conditions, varied loading directions, and various contingency situations. The results of CNN are compared with different machine learning techniques and tested in the IEEE 30 bus system and IEEE 118 bus system under various operational circumstances. The results illustrate the efficiency of the suggested method for online assessment of a power system's Voltage Stability Margin. The proposed approach gives real-time insights into voltage stability margins by leveraging the power of data and powerful algorithms, allowing grid operators to proactively manage and optimize renewable energy integration while ensuring grid dependability and stability.

SA-SRYOLOv8: A Research on Star Anise Variety Recognition Based on a Lightweight Cascaded Neural Network and Diversified Fusion Dataset

SA-SRYOLOv8: A Research on Star Anise Variety Recognition Based on a Lightweight Cascaded Neural Network and Diversified Fusion Dataset

Lightweight multi-scale feature dense cascade neural network for scene understanding of intelligent autonomous platform

Understanding the surrounding environment is an essential part of intelligent autonomous platform to complete autonomous driving or other autonomous tasks. Thereinto, scene understanding is a key technology to understand environmental information. For intelligent autonomous platform, it not only requires high accuracy, but also has a strict requirement for the inference speed. Moreover, it should be able to perform normally in complex environments such as target occlusion, the influence of illumination, and objects at different sizes. For tackling these problems, in this study, a lightweight multi-scale feature dense cascade neural network (LMFDCNet) is proposed in real-time scene understanding in complex scenes. Furthermore, trilateral fusion module is adopted to enhance low-level detailed features or high-level semantic features of each branch to describe the object better in different environments. A multi-scale upsampling module that fuses contextual information at different scales allows objects to be represented at different size. The results indicate that LMFDCNet produces 99.1 % classification accuracy on the tobacco dataset. On the Cityscapes dataset, we achieve 74.2 % Mean Intersection over Union (MIoU) with 13.6 M parameters at the speed of 38.2 FPS on a single 1070Ti card. To test the performance on the real-time application in the real world, LMFDCNet is deployed on the autonomous platform for real-time vision task such as classification of tobacco leaf state in curing process and semantic segmentation in the driving process of autonomous platform. The results reveal that LMFDCNet perform well on the AI embedded device and could be applied into intelligent autonomous platform.

Surface soil moisture from combined active and passive microwave observations: Integrating ASCAT and SMAP observations based on machine learning approaches

The fusion of active and passive microwave measurements is expected to provide more robust surface soil moisture (SSM) mapping across various environmental conditions compared to the use of a single sensor. Thus, the integration of the newest L-band passive (i.e., Soil Moisture Active Passive, SMAP) and the active (i.e., the Advanced Scatterometer, ASCAT) observations provides an opportunity for SSM mapping with improved accuracy. However, this integration remains largely underexplored. In this context, the integration of SMAP brightness temperature (TB) and ASCAT backscattering coefficients for global-scale SSM estimation was investigated, by fully considering the potential error sources in conventional radiative transfer models (RTMs) as well as other SSM linked factors. Based on ground measurements from globally distributed dense networks with mitigated mismatch issues and spatial/temporal independent evaluation strategies, this study: (i) comprehensively evaluated four classical machine learning approaches, including Random Forest (RF), Long-Short Term Memory (LSTM), Support Vector Machine (SVM), and Cascaded Neural Network (CNN), and chose the best performing RF method to implement the final integration of SSM; (ii) compared the integration retrievals to those made using data from a single sensor (SMAP or ASCAT) with the same machine learning framework, as well as to the SMAP passive, ASCAT active, and ESA CCI active-passive combined SSM products. The results show the integration retrievals achieve satisfactory performance by obtaining an averaged unbiased root mean squared error (ubRMSE) of 0.042 m3/m3 and a temporal correlation of 0.756, which are superior to machine learning based SSM estimated from a single active or passive sensor, and also outperform the SMAP, ASCAT, and ESA CCI products. Moreover, the temporal resolution is evidently improved compared to the SMAP and ASCAT SSM products, with a temporal ratio exceeding 60% for most areas across the globe. Therefore, blending active and passive measurements affords a more reliable SSM mapping with increased sampling at the global scale, and could contribute to improved hydro-ecological applications.

Performance evaluation of advanced machine learning methodologies in simulating hydrogen chloride (HCl) absorption by deep eutectic solvents

This work aims to construct the physical-based machine learning (ML) approach for estimating the hydrogen chloride (HCl) solubility in eleven different deep eutectic solvents (DESs). All DES-HCl equilibrium records versus temperatures and pressures are gathered from the literature to ensure ML universality. Then, a well-known feature selection technique was applied to identify the physically meaningful descriptors for the hydrogen bond acceptors (HBA) and hydrogen bond donors (HBD). Afterward, temperature, pressure, selected meaningful physical descriptors, and HBA/HBD molar concentrations are utilized to design ML tools to compute HCl solubility in DESs. Diverse MLs, including extreme learning machines, convolutional neural networks, long short-term memory, cascade neural networks, and extreme gradient boosting (XGBoost), are tested, and their accuracy is compared to select the best candidate. The XGBoost model predictions for HCl solubility in DESs are in acceptable agreement with the literature records. This model precisely simulates 417 DES-HCl equilibrium records in the temperature range of 293–348 K and pressures of up to 127 kPa with a correlation coefficient of 0.99621, mean absolute percentage deviation of 3.04, mean absolute error of 0.448, and root mean square error of 0.729. Results of the designed XGBoost model prove that acetamide-ethylene glycol (1:1), which can averagely dissolve up to 32.47 wt% (weight percent) of HCl, is the best DES for the HCl capture and recovery. After approving the XGBoost model reliability with different visual and numerical inspections, the Leverage method also justified its validity domain by more than 97.6%.

Based on neural network cascade abnormal texture information dissemination of classification of patients with schizophrenia and depression

This study used MRI brain image segmentation to identify novel magnetic resonance imaging (MRI) biomarkers to distinguish patients with schizophrenia (SCZ), major depressive disorder (MD), and healthy control (HC). Brain texture measurements, including entropy and contrast, were calculated to capture variability in adjacent MRI voxel intensity. These measures are then applied to group classification in deep learning techniques and combined with hierarchical correlations to locate results. Texture feature maps were extracted from segmented brain MRI scans of 141 patients with schizophrenia (SCZ), 103 patients with major depressive disorder (MD) and 238 healthy controls (HC). Gray scale coassociation matrix (GLCM) is a monomer matrix calculated in a voxel cube. Deep learning methods were evaluated to determine the application capability of texture feature mapping in binary classification (SCZ vs. HC, MD vs. HC, SCZ vs. MD). The method is implemented by repeated nesting and cross-validation for feature selection. Regions that show the highest correlation (positive or negative). In this study, the authors successfully classified SCZ, MD and HC. This suggests that texture analysis can be used as an effective feature extraction method to distinguish different disease states. Compared with other methods, texture analysis can capture richer image information and improve classification accuracy in some cases. The classification accuracy of SCZ and HC, MD and HC, SCZ and MD reached 84.6%, 86.4% and 76.21%, respectively. Among them, SCZ and HC are the most significant features with high sensitivity and specificity.

Active disturbance rejection control of flexible industrial manipulator: A MIMO benchmark problem

Robotic manipulators are widely used in modern industry. In order for robot manipulators to achieve trajectory tracking control, the end-effectors must move precisely along the given trajectories. The nonlinearity, strong coupling, and uncertainty of the system make trajectory tracking very complicated and difficult. An approach for robust feedback control is presented in this paper as a solution to this problem. The design relies on active disturbance rejection control to estimate unknown disturbances affecting the plant. The controlled system is an uncertain two-link manipulator with elastic gear transmissions, described by nonlinear friction and elasticity. It is based on an industrial benchmark problem. The model has two significant uncertainties: parametric uncertainty as well as external disturbances that affect tools and motors. A simulation-based evaluation of the proposed control method and a comparison of its performance, considering specific robustness requirements and trajectory tracking, are presented. The comparison includes the baseline controller, the QFD (Quality Function Deployment) method, and PD-CNN (PD Control Compensation based on the Cascade Neural Network). Computer simulations of the proposed controller are carried out to validate the ABB (Asea Brown Boveri) industrial benchmark. The simulation results demonstrate superiority in meeting stability and target value requirements. The stability and ultimate boundedness of the tracking error of ADRC (Active Disturbance Rejection Control) are demonstrated.

Real-Time Noise Suppression Method Using Deep Learning Algorithm

In this study, we present an innovative approach to real-time facial feature detection utilizing the MTCNN (MultiTask Cascaded Neural Network) deep learning architecture. Unlike traditional methods, our novel framework combines advanced techniques to achieve unparalleled precision and efficiency in facial feature localization. Through a meticulous exploration of MTCNN's capabilities, we unveil a transformative methodology that significantly enhances the speed and accuracy of real-time facial detection. Our research focuses on pushing the boundaries of existing facial recognition technologies, introducing a fresh perspective on the application of MTCNN. By leveraging its unique architecture, we not only address the challenges associated with real-time detection but also enhance the overall robustness of the system. The proposed approach showcases the untapped potential of MTCNN, establishing it as a key player in the realm of facial feature detection. Through rigorous experimentation and evaluation, we demonstrate the superiority of our approach over conventional methods, highlighting its effectiveness in diverse scenarios. This work contributes to the ongoing evolution of deep learning applications in computer vision, with implications for security, surveillance, and various human-computer interaction domains. Our findings open new avenues for researchers and practitioners seeking cutting-edge solutions in the dynamic field of real-time facial feature detection.

PlantBiCNet: A new paradigm in plant science with bi-directional cascade neural network for detection and counting

Deep learning is increasingly popular for precise plant detection and counting in computer vision applications. Despite the rising cross-disciplinary research in this field, an important issue has been neglected: due to the inherent constraints imposed by the natural growth patterns of plants, the costly and time-consuming process of collecting plant image data, resulting in a scarcity of real-world available data. Consequently, researchers are often constrained to exploring large-scale data representations or limited applications. Currently, still no generalized network architecture excels in demanding environments. To address this, we introduce two new bi-directional cascade neural network paradigms, PlantBiCNet and its lightweight variant, PlantBiCNet-Lite. Unlike previous research, we place greater emphasis on the network decoding process itself rather than the encoder. Our bi-directional cascade and weight fusion decoding approach maximizes high-level semantics and low-level spatial data, enhancing the localization signal in a way that respects gradient diversity. We evaluate PlantBiCNet and PlantBiCNet-Lite on five public plant detection and counting datasets and a novel plant dataset. The average precision AP50 on these six small-scale datasets reached 0.905, with a high coefficient of determination R2 of 0.935, while the reported frame rate was 196 FPS. These results significantly outperform state-of-the-art computer vision methods. Moreover, we found that the introduction of the Dropout mechanism markedly enhances the model's performance and generalization ability. Further analysis supports our experiments and provides valuable insights into network optimization. Overall, our organization's open datasets and network modeling details provide a solid foundation for future research in the field of plant science.

Stainless steel cylindrical pot outer surface defect detection method based on cascade neural network

Addressing issues such as surface geometric distortion, high reflection, and the challenge of detecting minor defects on the external surfaces of stainless steel pots, this paper presents a novel approach for detecting defects on the external surfaces of cylindrical stainless steel pots based on high-resolution line-scan imaging. The method begins by creating a real-time inspection system that includes a line-scan camera and a strip light source. This system achieves distortion-free, high-resolution image acquisition of the outer surfaces of stainless steel cylindrical pots by establishing linear constraints on pot size, rotary table rotation speed, and line-scan frame rate. Subsequently, a You Only Look Once and fully convolutional network cascade neural network surface defect detection strategy is introduced, utilizing dual-channel images of the original and enhanced images as inputs. This approach enables the characterization of subtle defects in high-resolution image data. Finally, we conducted experiments using the proposed method on the provided dataset, and the results demonstrate the effectiveness of this approach in detecting various types of product defects. The mean pixel accuracy achieved a remarkable 91.69%, while the mean intersection over union score reached an impressive 83.59%. These findings provide an effective technical means for the qualitative detection of the types of defects on the surface of stainless steel pots and the quantitative measurement of the size of the defects.

Simultaneous retrieval of aerosol and ocean properties from PACE HARP2 with uncertainty assessment using cascading neural network radiative transfer models

Abstract. The University of Maryland, Baltimore County (UMBC) Hyper-Angular Rainbow Polarimeter (HARP2) will be on board NASA's Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission, scheduled for launch in January 2024. In this study we systematically evaluate the retrievability and uncertainty of aerosol and ocean parameters from HARP2 multi-angle polarimeter (MAP) measurements. To reduce the computational demand of MAP-based retrievals and maximize data processing throughput, we developed improved neural network (NN) forward models for spaceborne HARP2 measurements over a coupled atmosphere and ocean system within the FastMAPOL retrieval algorithm. To this end, a cascading retrieval scheme is implemented in FastMAPOL, which leverages a series of NN models of varying size, speed, and accuracy to optimize performance. Two sets of NN models are used for reflectance and polarization, respectively. A full day of global synthetic HARP2 data was generated and used to test various retrieval parameters including aerosol microphysical and optical properties, aerosol layer height, ocean surface wind speed, and ocean chlorophyll a concentration. To assess retrieval quality, pixel-wise retrieval uncertainties were derived from error propagation and evaluated against the difference between the retrieval parameters and truth based on a Monte Carlo method. We found that the fine-mode aerosol properties can be retrieved well from the HARP2 data, though the coarse-mode aerosol properties are more uncertain. Larger uncertainties are associated with a reduced number of available viewing angles, which typically occur near the scan edge of the HARP2 instrument. Results of the performance assessment demonstrate that the algorithm is a viable approach for operational application to HARP2 data after the PACE launch.

Spatial Localization of Electromagnetic Radiation Sources by Cascade Neural Network Model with Noise Reduction

Spatial Localization of Electromagnetic Radiation Sources by Cascade Neural Network Model with Noise Reduction

Real-Time Cascading Failure Risk Evaluation With High Penetration of Renewable Energy Based on a Graph Convolutional Network

When renewable energy penetration increases and the power network becomes more complex, vulnerability and security concerns typically also become more prevalent. A proper evaluation method focused on risk prediction and failure identification is required in modern power systems. In our work, a novel evaluation framework with high efficiency and accuracy is proposed, where the influence of over/under-voltage on renewable energy generators is considered during the cascading failure risk evaluation process. A graph convolutional network and a long short-term memory network are applied to describe the complex network topological structure and extract the system's electrical features, respectively. The neural network model trained with offline simulation data can reduce the numerical effort of evaluating cascading failure risk in real-time. Finally, the proposed method is verified based on the benchmark of both an IEEE 39-bus test system and a real-world French power system. The results show that the proposed cascade neural network model can achieve better evaluation results with shorter training and evaluation times.

Multiple-instance ensemble for construction of deep heterogeneous committees for high-dimensional low-sample-size data

Deep ensemble learning, where we combine knowledge learned from multiple individual neural networks, has been widely adopted to improve the performance of neural networks in deep learning. This field can be encompassed by committee learning, which includes the construction of neural network cascades. This study focuses on the high-dimensional low-sample-size (HDLS) domain and introduces multiple instance ensemble (MIE) as a novel stacking method for ensembles and cascades. In this study, our proposed approach reformulates the ensemble learning process as a multiple-instance learning problem. We utilise the multiple-instance learning solution of pooling operations to associate feature representations of base neural networks into joint representations as a method of stacking. This study explores various attention mechanisms and proposes two novel committee learning strategies with MIE. In addition, we utilise the capability of MIE to generate pseudo-base neural networks to provide a proof-of-concept for a “growing” neural network cascade that is unbounded by the number of base neural networks. We have shown that our approach provides (1) a class of alternative ensemble methods that performs comparably with various stacking ensemble methods and (2) a novel method for the generation of high-performing “growing” cascades. The approach has also been verified across multiple HDLS datasets, achieving high performance for binary classification tasks in the low-sample size regime.

Tunnel Lining Crack Recognition Algorithm Integrating SK Attention and Cascade Neural Network

Cracks are one of the main types of tunnel-lining defects. At present, there are no particularly good methods for identifying tunnel-lining cracks. The methods that are used are associated with problems such as poor robustness, low detection efficiency, and inconsistency in defect identification. Vision-based crack identification algorithms that use images of tunnel-lining cracks are also affected by problems such as weak features, irregular shapes, random development, and small proportions. In light of this, we propose in this paper a neural network that integrates U-net and FPN (feature pyramid network) pyramid structure characteristics. Because of its ability to output deep feature information, an SK (selective kernel) attention mechanism was integrated to increase the weight of effective feature regions and improve the detection accuracy of feature maps. An RPN (region proposal network) regional detection network was used for multi-scale regional screening, and non-maximum suppression was used to eliminate repeated anchor frames between different regions. The continuity of cracks was determined using adaptive regional expansion. The region segmentation network was also combined with multi-scale feature maps. Finally, the segmentation results for regions of fine features could be mapped to the whole image to identify cracks, thereby solving the problem that arises when textural features of cracks are weak and cannot be accurately identified. For testing purposes, photos of tunnel-lining crack defects were used. The algorithm described in this paper was able to combine deep and shallow features to identify more abundant crack features, and the recognition accuracy of the crack classification network reached 98.51%. With respect to crack segmentation, the algorithm’s segmentation accuracy reached 94.55%, and its single processing time reached 60.76 ms, indicating more accurate and efficient segmentation performance, compared with FCNs (fully convolutional networks) and classical U-net networks.

Elman and cascade neural networks with conjugate gradient Polak-Ribière restarts to predict diesel engine performance and emissions fueled by butanol as sustainable biofuel

This study explores the possibility of utilizing green butanol, a promising type of biofuel, in diesel engines to measure its impact on engine performance and emissions. Experimental data were collected on several parameters, including brake-specific fuel consumption (BSFC), brake thermal efficiency (BTE), exhaust gas temperature, all of which are influenced by the biofuel type used, in this case, butanol. The study examined emissions such as carbon monoxide (CO), hydrocarbons (HC), nitrogen oxides (NOx), and smoke opacity. Such emissions have the potential to improve with the use of different biofuels ratios, such as butanol. Advanced machine learning techniques, Elman and Cascade Neural Networks, were employed to predict the performance and emission characteristics of engines using butanol. The models were trained using a Conjugate Gradient Learning Function with Polak-Ribière Restarts to simulate the effects of butanol as biofuel, on diesel engines. Key findings revealed that when incorporating butanol into diesel fuel blends, potential improvements in BTE and fuel efficiency were observed. Notably, using butanol as a biofuel reduced exhaust gas temperatures and CO emissions, demonstrating the potential of this particular biofuel. Conversely, there were observed increases in HC emissions and smoke opacity, signifying the complexities of using biofuels such as butanol. Cascade neural network proved to be highly accurate in predicting engine performance parameters fueled with butanol as biofuel. Overall, the study offers valuable insights into the use of butanol as a biofuel, its potential benefits, and challenges, underscoring the importance of continuous research in sustainable biofuels such as butanol.

Risk-averse and flexi-intelligent scheduling of microgrids based on hybrid Boltzmann machines and cascade neural network forecasting

The future of energy flexibility in microgrids (MGs) is steering towards a highly granular control of the end-user customers. This calls for more highly accurate uncertainty forecasting and optimal management of risk and flexibility options. This paper presents a novel data-driven model to optimize the operation of MGs based on a risk-averse flexi-intelligent energy management system (RFEMS), considering the rising challenge of global climate change. It considers the presence of renewables, a diesel generator, and flexibility resources (FRs) containing a demand response program (DRP), distributed electric vehicles (EVs), and electric springs (ESs). In the first phase, the proposed model, by means of a novel hybrid deep-learning (DL) model, forecasts uncertain parameters associated with wind and solar generations, load demand, and day-ahead energy market price. The architecture of the proposed hybrid forecasting model is composed of several stacked restricted Boltzmann machines and a cascade neural network. In the second phase, the MG operator (MGO), based on the obtained uncertainty forecasting results, in the context of a hybrid risk-controlling model, optimizes the MG operation using the provided demand-side flexibility. The proposed optimization problem is linearized stochastic programming with robust concepts, subject to AC optimal power flow constraints, MG flexibility restrictions, and operating limits of local resources. Finally, the efficiency of the proposed RFEMS by using real German datasets on a 33-bus test MG is analyzed. Numerical results demonstrate the superior performance of the proposed forecasting model over several hybrid DL models. In particular, the root mean square error (RMSE) index for wind, solar, load, and price forecasting is improved by 53.35%, 73.24%, 80.24%, and 58.1%, respectively. Further analysis of the proposed RFEMS reveals that operating indices in two 33-bus and 69-bus test networks are significantly improved. It paves pathways to risk-averse, flexible, and economic operation of smart active distribution networks.

Cross-domain fashion cloth retrieval via novel attention-guided cascade neural network and clothing parsing

Cross-domain fashion retrieval is a feasible method for finding the required clothing among a large number of products. Computer vision-based retrieval algorithms have been developed by many researchers. However, the precise retrieval process still faces challenges such as corrupted query images, ambiguous patterns, and an array of attributes. In this work, a novel attention-guided cascaded network has been proposed for efficient cloth retrieval by overcoming the aforementioned challenges. Initially, the input images are pre-processed with Wavelet integrated Retinex algorithm (WRA) to enhance the query cloth image without diminishing the important information. The attention-guided cascaded network is designed with global feature descriptor (GFD) and local feature descriptor (LFD) modules for extracting the domain-enriched features. These two modules are integrated with an attention block to reduce complexity and improve performance using selected features. In the LFD module, the Clothing Parsing Encoder-Decoder network (CPED-Net) is used to separate the bottom and top regions and sends them into the respective network for fine-grained feature analysis. Finally, global and local features from the modules are combined in similarity analysis for cloth retrieval. The proposed model achieves an overall accuracy of 98.54% based on the fashion retrieval benchmarks dataset. From the experimental results, the proposed model is superior to the state-of-the-art models.

A promising artificial intelligence-based tool to simulate the efficient and sustainable hydrogen sulfide elimination using deep eutectic solvents

This work applies artificial intelligence (AI) methodologies and relevance tests to investigate the hydrogen sulfide (H2S) elimination capacity of a wide range of DESs (deep eutectic solvents). A great deal of effort is made to deploy a practical model to predict the H2S removal ability of eighteen different DESs as a function of DES chemistry, equilibrium conditions, and H2S-DES interaction. Multiple linear regression and Pearson’s correlation methods are employed to measure the direction and relative influence of hydrogen bond acceptor (HBA), hydrogen-bond donor (HBD), the HBA and HBD concentration, water dose, pressure, temperature, and absorption enthalpy on the DES capacity to absorb H2S molecules. Both tests approve that absorption enthalpy (ΔH) of H2S by DESs, hydrogen bond acceptor (HBA), pressure, and temperature mainly control the considered separation process. The relevance tests also identify that the H2S solubility in DESs improves by either a decrease in ΔH, HBA molecular weight, and temperature or an increase in pressure. Then, five diverse AI types are applied to estimate H2S molality in DESs as a function of the involved explanatory variables, and their accuracy is compared. The sensitivity analysis clarifies that the cascade neural network (CNN) with only 20 hidden nodes is the most reliable AI tool to simulate the considered problem. Further investigations approve that the CNN tool provides more accurate predictions for H2S solubility in DESs when it is equipped with the logarithm and tangent sigmoid transfer functions in the hidden and output neuronic layers. Also, amongst eleven checked training algorithms, the Levenberg-Marquardt presented the best performance. This structure-regulated CNN tool can predict 512 actual H2S solubility records in DESs with the MSE = 0.0031, MAE = 0.02, RAE = 2.03%, AARE = 3.03%, and multiple-R = 0.99943.