Artificial Neural Network Topology Research Articles

Sparsity-Aware In-Memory Neuromorphic Computing Unit With Configurable Topology of Hybrid Spiking and Artificial Neural Network

Sparsity-Aware In-Memory Neuromorphic Computing Unit With Configurable Topology of Hybrid Spiking and Artificial Neural Network

Performance Prediction of A Power Generation Gas Turbine Using An Optimized Artificial Neural Network Model

This paper presents the application of an Artificial Neural Network (ANN) based model for performance ‎prediction of a power generation gas turbine. The suggested model was optimized to provide a large ‎database for comparison between different ANN topologies. Then, based on the optimization results, the ‎Multi-Layer Perceptron (MLP) of two layers was constructed and utilized for this study as the best-‎optimized topology. Training of this model was done using historical operational data of a Rolls Royce ‎‎(RB21-24G) gas turbine unit. The outcome results from this model used for performance prediction show ‎good accuracy for different ambient conditions and variable power ratings. Then, a degradation study was ‎also introduced comparing measurements of the same gas turbine utilizing one year later, on-site ‎operational data, with the predicted values generated by the ANN model. The result shows consistency ‎between the measured data and the model results.‎

Elegans-AI: How the connectome of a living organism could model artificial neural networks

This paper introduces Elegans-AI models, a class of neural networks that leverage the connectome topology of the Caenorhabditis elegans to design deep and reservoir architectures. Utilizing deep learning models inspired by the connectome, this paper leverages the evolutionary selection process to consolidate the functional arrangement of biological neurons within their networks. The initial goal involves the conversion of natural connectomes into artificial representations. The second objective centers on embedding the complex circuitry topology of artificial connectomes into both deep learning and deep reservoir networks, highlighting their neural-dynamic short-term and long-term memory and learning capabilities. Lastly, our third objective aims to establish structural explainability by examining the heterophilic/homophilic properties within the connectome and their impact on learning capabilities. In our study, the Elegans-AI models demonstrate superior performance compared to similar models that utilize either randomly rewired artificial connectomes or simulated bio-plausible ones. Notably, these Elegans-AI models achieve a top-1 accuracy of 99.99% on both Cifar10 and Cifar100, and 99.84% on MNIST Unsup. They do this with significantly fewer learning parameters, particularly when reservoir configurations of the connectome are used. Our findings indicate a clear connection between bio-plausible network patterns, the small-world characteristic, and learning outcomes, emphasizing the significant role of evolutionary optimization in shaping the topology of artificial neural networks for improved learning performance.

Optimising Hyperparameters of Artificial Neural Network Topology for SHM Damage Detection and Identification

Optimising Hyperparameters of Artificial Neural Network Topology for SHM Damage Detection and Identification

ANN based smart sensing of short-term voltage stability assessment of modern power system

This study addresses the critical challenge of Short-term Voltage Stability (STVS) in the contemporary electric grid system, characterized by diverse static loads, induction motors, electronic loads, electric vehicle charging stations, and renewable energy sources. High-power sensing elements are identified for power system stability and transient measurements, and the existing system is explored for fault measurement and stability maintenance. The analysis employs an IEEE 14-bus dynamic model with two distinct load configurations. In the first configuration, the static load at bus 14 is replaced with a composite load, encompassing static loads, induction motors, electronic loads, and EV charging stations. For the second novel configuration at bus 14, the static load is substituted with a composite load and a solar photovoltaic (PV) generator operating in reactive power compensation mode. This addition of the PV generator compensates for elevated reactive power demand in the 14-bus system. The study conducts voltage stability analysis in two phases, evaluating the voltage profile and transmission line load ability in both scenarios during the first phase. In the second phase, STVS is simulated by briefly opening the circuit breaker in bus 2. The results reveal notable improvements in the voltage profile of all buses, a 6 % enhancement in power factor at bus 14 in the dynamic model with the novel configuration, and a 3 % increase in the loading factor of the IEEE Bus 14 system. The dynamic model incorporating a PV generator and composite load in bus 14 demonstrates superior STVS performance compared to the setup with only a composite load. The abstract concludes by highlighting the development of real-time voltage profile assessment using Artificial Neural Network (ANN) topology, leveraging results from the Continuous Power Flow (CPF) analysis of the novel model with a PV generator and composite load in the far end bus. This restructuring ensures a logical and clear organization, encompassing an introduction to the problem, a description of methods employed, presentation of results, and a succinct conclusion summarizing the key findings.

Adaptive Neuroevolution With Genetic Operator Control and Two-Way Complexity Variation

Topology and weight evolving artificial neural network (TWEANN) algorithms optimize the structure and weights of artificial neural networks (ANNs) simultaneously. The resulting networks are typically used as policy models for solving control and reinforcement learning (RL) type problems. This paper presents a neuroevolution algorithm that aims to address the typical stagnation and sluggish convergence issues present in other neuroevolution algorithms. These issues are often caused by inadequacies in population diversity preservation, exploration/exploitation balance, and search flexibility. This new algorithm, called the Adaptive Genomic Evolution of Neural-Network Topologies (AGENT), builds on the neuroevolution of augmenting topologies (NEAT) concept. Novel mechanisms for adapting the selection and mutation operations are proposed to favorably control population diversity and exploration/exploitation balance. The former is founded on a fundamentally new way of quantifying diversity by taking a graph-theoretic perspective of the population of genomes and inter-genomic differences. Further advancements to the NEAT paradigm occur through the incorporation of variable neuronal properties and new mutation operations that uniquely allow both the growth and pruning of ANN topologies during evolution. Numerical experiments with benchmark control problems adopted from the OpenAI Gym illustrate the competitive performance of AGENT against standard RL methods and adaptive HyperNEAT, and superiority over the original NEAT algorithm. Further parametric analysis provides key insights into the impact of the new features in AGENT. This is followed by evaluation on an unmanned aerial vehicle collision avoidance problem where maneuver planning models are learnt by AGENT with 33% reward improvement over 15 generations.

Modelling selenite and chemical oxygen demand removal in an inverse fluidized bed bioreactor using genetic algorithm optimized artificial neural network

AbstractBACKGROUNDA genetic algorithm (GA) optimized artificial neural network (ANN) model was developed to simulate selenite removal efficiency (SRe) and chemical oxygen demand (COD) removal efficiency (CRe) in a single‐stage inverse fluidized bed reactor (IFBR) operated under different hydraulic retention time (HRT) (12–48 h), and inlet selenite concentration (12.7–635 mg L−1). Multi‐objective optimization for SRe and CRe with the ANN model was adopted to estimate the pareto optimal solutions in the bioreactor system.RESULTSGA optimized ANN topology suggested five neurons in the hidden layer, and the optimal data partitioning between training, test, and validation subsets (in 70:15:15 ratio) using cross‐validation. The average fitness for SRe and CRe of the resulting ANN model over training (R2 = 0.885), test (R2 = 0.967) and validation (R2 = 0.972) subsets were reasonably good. Sensitivity analysis suggested an increase in SRe and CRe with HRT, while an increase in inlet selenite concentration had a negative effect on SRe. Local interpretable model‐agnostic explanations and SHapley Additive exPlanations for the developed model suggested a positive coefficient of the predictors on CRe. Pareto optimal solution for SRe (92.87%–95.26%), and CRe (91.33%–99.35%) were suggested for HRT and inlet selenite concentration varying between 35.67 h and 36 h, and 0.1 g L−1 and 0.884 g L−1, respectively.CONCLUSIONSThe GA optimized ANN model could be exploited for the prediction of contaminant removal efficiencies in the bioreactor system with a fair degree of complexity (i.e., IFBR). Multiple objective pareto optimal solutions can identify the best operating conditions relevant for wastewater treatment. © 2023 Society of Chemical Industry (SCI).

Multi-objective optimization of breakthrough times for hydrogen purification through layered bed pressure swing adsorption based on genetic algorithm and artificial neural network model

Hydrogen purification from steam methane reforming (SMR) by pressure swing adsorption (PSA) technology is a common method to obtain high-purity hydrogen. The breakthrough time can well reflect the adsorption dynamics and help PSA cycle design. In order to avoid time-consuming and labor-intensive breakthrough curve experiments, it is very necessary to develop a fast and accurate surrogate model. In this study, a genetic algorithm (GA) and artificial neural network (ANN) are combined to predict and optimize breakthrough times of adsorbates in activated carbon/zeolite layered beds. Using the Latin hypercube sampling strategy, training data sets of GA-optimized ANN (GA-ANN) obtains from the physical model of adsorption, heat and mass transfer model. The genetic algorithm (GA) optimizes the weights and biases of ANN for better performance. The ANN topology has 4 input variables (superficial velocity, activated carbon height, adsorption pressure and feed temperature) and 3 output variables (breakthrough times of CH4, CO and CO2). The number of neurons in the hidden layer for the GA-ANN model was optimized as 7 to predict and optimize the maximum breakthrough time of SMR with a four-component (H2/CH4/CO/CO2) system. The sensitivity analysis displays that relative importance to the breakthrough time is in the order of superficial velocity (40.88%) > adsorption pressure (24.55%) > activated carbon height (21.04%) > feed temperature (13.53%). To obtain high-purity hydrogen, the GA-ANN model combined with multi-objective GA optimization is used to maximize the breakthrough times of CH4 and CO. The GA-ANN surrogate model proposed in this paper can not only accurately and quickly achieve the purpose of predicting the system breakthrough time with a high correlation coefficient (R = 0.9937) but also obtains the optimal operating conditions of PSA hydrogen purification.

ICA-LightGBM Algorithm for Predicting Compressive Strength of Geo-Polymer Concrete

The main goal of the present study is to investigate the capability of hybridizing the imperialist competitive algorithm (ICA) with an intelligent, robust, and data-driven technique named the light gradient boosting machine (LightGBM) to estimate the compressive strength of geo-polymer concrete (CSGCo). The hyper-parameters of the LightGBM algorithm have been optimized based on ICA and its accuracy improved. The obtained results from the proposed hybrid ICA-LightGBM are compared with the traditional LightGBM model as well as four different topologies of artificial neural networks (ANN) comprising a multi-layer perceptron neural network (MLP), radial basis function (RBF), generalized feed-forward neural network (GFFNN), and Bayesian regularized neural network (BRNN). The results of these models were compared based on three evaluation indices of R2, RMSE, and VAF for providing an objective evaluation of the performance and capability of the predictive models. Concerning the outcomes, the ICA-LightGBM with the R2 of (0.9871 and 0.9805), RMSE of (0.4703 and 1.3137), and VAF of (98.5773 and 98.0397) for training and testing phases, respectively, was a superior predictor to estimate the CSGCo compared to the LightGBM with the R2 of (0.9488 and 0.9478), RMSE of (0.9532 and 2.1631), and VAF of (94.3613 and 94.5173); the MLP with the R2 of (0.9067 and 0.8959), RMSE of (1.3093 and 3.3648), and VAF of (88.9888 and 84.9125); the RBF with the R2 of (0.8694 and 0.8055), RMSE of (1.4703 and 5.0309), and VAF of (86.3122 and 66.1888); the BRNN with the R2 of (0.9212 and 0.9107), RMSE of (1.1510 and 2.6569), and VAF of (91.4168 and 90.5854); and the GFFNN with the R2 of (0.9144 and 0.8925), RMSE of (1.1525 and 2.9415), and VAF of (91.4092 and 88.9088). Hence, the proposed ICA-LightGBM algorithm can be efficiently used in anticipating the CSGCo.

Artificial neural networks based on segmented model for adaptability and stability evaluation of soybean genotypes

Unlike models based on simple linear regressions, segmented models can better assess the adaptability and stability of genotypes, demonstrating a nonlinear response pattern over environmental variation. However, these methods can be under statistical limitations, such as the Type Error II increase and biased estimates. Therefore, this work aimed to transpose the concepts of adaptability and stability from the statistical analysis of a segmented model to the discriminatory potential of an artificial neural network (ANN) and use it to classify soybean (Glycine max (L.) Merr.) genotypes. An ANN training was carried out with the grain yield of 7,200 soybean genotypes simulated in 15 different environments. The ANN topology chosen was the one that had less than 1% of error in the testing phase with 1,800 simulated genotypes. A total of 9,000 simulated soybean genotypes were previously arranged in 18 different classes, which represented the combination of nine classes of adaptability by the Verma and collaborators (VCM) method and two classes of stability (invariability concept) by the Finlay & Wilkinson (FW) method. Finally, the grain production of ten real soybean genotypes was inputted into the ANN-trained model, and the classification regarding adaptability and stability was obtained. There was 90% agreement between the ANN and VCM analyses regarding the adaptability classification and 20% regarding stability. With the methods presented in this work, it was demonstrated that the potential of using ANNs to assess the adaptability of genotypes is strong. In addition, since stability was introduced in the ANN as a different concept from that used to classify the genotypes by the statistical method, such classification needs to be reviewed and further improved

Predicción del tiempo de almacenamiento de carne congelada usando modelado de redes neuronales artificiales con valores de color

Among the various methods available to determine the storage time of frozen meat, including analyses based on physical and chemical properties, sensory analysis, particularly color changes, is an important aspect of meat acceptability for consumers. In this study, an artificial neural network (ANN) was employed to predict the storage time of the meat based on the CIELAB color space, represented by the Lab* (L*), (a*), and (b*) values measured by a computer vision system at two–month intervals over a period of up to one year. The ANN topology was optimized based on changes in correlation coefficients (R2) and mean square errors (MSE), resulting in a network of 60 neurons in a hidden layer (R2 = 0.9762 and MSE = 0.0047). The ANN model's performance was evaluated using criteria such as mean absolute deviation (MAD), MSE, root mean square error (RMSE), R2, and mean absolute error (MAE), which were found to be 0.0344, 0.0047, 0.0687, 0.9762, and 0.0078, respectively. Overall, these results suggest that using a computer vision–based system combined with artificial intelligence could be a reliable and nondestructive technique for evaluating meat quality throughout its storage time.

Detection of Malicious Network Activity by Artificial Neural Network

This paper presents a deep learning approach to detect malicious communication in a computer network. The intercepted communication is transformed into behavioral feature vectors that are reduced (using principal component analysis and stepwise selection methods) and normalized to create training and test sets. A feed-forward artificial neural network is then used as a classifier to determine the type of malicious communication. Three training algorithms were used to train the neural network: the Levenberg-Marquardt algorithm, Bayesian regularization, and the scaled conjugate gradient backpropagation algorithm. The proposed artificial neural network topology after reducing the size of the training and test sets achieves a correct classification probability of 81.5 % for each type of malicious communication and of 99.6 % (and better) for normal communication.

Investigation of best artificial neural network topology to model the dynamic viscosity of MWCNT-ZnO/SAE 5W30 nano-lubricant

Nanofluids (NFs) are another achievement of nanoscience that has received much attention in recent years. The properties of the base fluid are improved when MWCNT nanoparticles (NPs) are added, according to research. Due to their high efficiency, reliability, and flexibility, dynamic viscosity (μnf) is one of the crucial properties of NFs that affect heat transfer and surface tension. Due to their intelligent structure, artificial neural networks (ANNs), a novel modeling technique, compete with traditional statistical modeling. The correlation of μnf to three input parameters for MWCNT-ZnO/SAE 5W30 NF was studied, and input-output model was attained via modeling, and training an ANN. This ANN achieved R2 and MSE values of 0.9999 and 0.00010349, respectively, demonstrating the ANN's successful training. In this study, μnfshows a strong inverse effect with temperature and shear rate (SR). In other words, the temperature increases from 5 to 55 oC almost reduces μnffrom 370 cP to around 30 cP, and SR increases from 50 to 1000 rpm almost reduces μnffrom 400 cP to 25 cP. Additionally, the solid volume fraction (SVF) affected μnf directly although insignificantly.

Modern Artificial Neural Networks: Is Evolution Cleverer?

Machine learning tools, particularly artificial neural networks (ANN), have become ubiquitous in many scientific disciplines, and machine learning-based techniques flourish not only because of the expanding computational power and the increasing availability of labeled data sets but also because of the increasingly powerful training algorithms and refined topologies of ANN. Some refined topologies were initially motivated by neuronal network architectures found in the brain, such as convolutional ANN. Later topologies of neuronal networks departed from the biological substrate and began to be developed independently as the biological processing units are not well understood or are not transferable to in silico architectures. In the field of neuroscience, the advent of multichannel recordings has enabled recording the activity of many neurons simultaneously and characterizing complex network activity in biological neural networks (BNN). The unique opportunity to compare large neuronal network topologies, processing, and learning strategies with those that have been developed in state-of-the-art ANN has become a reality. The aim of this review is to introduce certain basic concepts of modern ANN, corresponding training algorithms, and biological counterparts. The selection of these modern ANN is prone to be biased (e.g., spiking neural networks are excluded) but may be sufficient for a concise overview.

STEM TAPERING OF Eucalyptus spp. USING DIFFERENT CONFIGURATIONS OF ARTIFICIAL NEURAL NETWORKS

The objective of this work was to test and evaluate different configurations of artificial neural networks (ANNs) for modeling tree stem taper in Eucalyptus spp. in strands in the microregion of Pirapora, Minas Gerais. The data used came from 8,410 Eucalyptus spp. at different speeds. The quantitative variables measured were: age, total height, diameter at the height of 1.30 m (dbh), diameter and height in different positions on the stem. The only qualitative variable measured was the clone. Four scenarios were evaluated: scenario 1 with Ht, dbh, hi, A and Clone inputs; scenario 2 with Ht, dbh, hi and Clone; scenario 3 with Ht, dbh, hi and A; and scenario 4 with Ht, dbh and hi. We tested different ANNs topologies of the Multilayer Perceptron type. The ANNs 102 (neurons in the hidden layer = 18; function = Exponential; algorithm = Rprop), 91 (neurons in the hidden layer = 19; function = Exponential; algorithm = Rprop), 13 (neurons in the hidden layer = 7; Function = Exponential; Algorithm = SCG) and 27 (neurons in the hidden layer = 6; function = Exponential; algorithm = Rprop) presented the best measures of statistical accuracy in training to predict the bottleneck in scenarios 1, 2, 3 and 4, respectively. The ANN 103 (neurons in the hidden layer = 19; function = Exponential; algorithm = Rprop) from scenario 1 presented good statistical results in the validation. Thus, the ANNs were efficient in predicting the diameter along the Eucalyptus spp stem.

Sous-vide processing of silver carp: Effect of processing temperature and cold storage duration on the microbial quality of the product as well as modeling by artificial neural networks.

Silver carp (Hypophthalmichthys molitrixi) was processed by sous-vide method at different temperatures (60, 65, 70, and 75°C). Then, the microbiological quality of the processed samples was monitored during cold storage (4°C) for 21 days. The target microorganisms were Enterobacteriaceae, Lactic Acid bacteria (LAB), Pseudomonas, Psychrotrophs, and total viable count (TVC). In samples processed at 75°C, the presence of Enterobacteriaceae, Pseudomonas and Psychrotrophs were not detectable up to 15 days of storage and lactic acid bacteria were not detectable even at the end of the storage period. A radial basis function neural network (RBFNN) model was established to predict the changes in the microbial content of silver carp. In this step, the relationship between processing temperature and storage duration on microbial growth was modeled by ANNs (artificial neural networks). The optimal ANN topology for modeling Enterobacteriaceae, Pseudomonas, and Psychrotroph contained 9 neurons in the hidden layer, but it contained 15 and 14 neurons for TVC and LAB, respectively. By experimenting with the temperature of -80°C, it was revealed that the obtained ANN model has a high potential for prediction.

An efficient hybrid LSTM-ANN joint classification-regression model for PPG based blood pressure monitoring

This paper investigates the importance of classification in optimizing the estimation accuracy of blood pressure (BP) using photoplethysmography (PPG) signal features, with the aim of balancing efficiency and performance. Accordingly, the experimental design of this study was employed by constructing various artificial neural network (ANN) and long short-term memory (LSTM) topologies such as conventional single-stage models and classification-regression schemes including two-stage models, joint learning models, and the hybrid LSTM-ANN model. All models were tested on 40 subjects from the Physionet’s MIMIC II database. With only 4 features, the proposed hybrid LSTM-ANN model achieved the best MAE ± SD of 3.39 ± 5.47 and 1.79 ± 3.72 mmHg for SBP and DBP, respectively. Furthermore, the single-stage and the two-stage ANN-based models were selected to be embedded in STM32 Microcontroller Unit (MCU) for real-time predictions of BP. The two-stage classification-regression ANN model showed superior performance and robustness in real-time testing on six subjects, achieved an MAE ± SD of 1.41 ± 1.29 mmHg, resulting in an 83.5% reduction of MAE compared with the traditional single-stage ANN model. Overall, the data presented in this work substantiate the significant role of classification in improving regression accuracy in both ANN and LSTM-based models. Compared to other state-of-the-art models, the hybrid LSTM-ANN with a joint classification-regression approach was found to be the optimal architecture to achieve the best performance and a trade-off between memory usage, complexity, and latency with acceptable BP prediction accuracy.

Estimation of surface soil moisture by combining a structural equation model and an artificial neural network (SEM-ANN)

Soil moisture is an important variable of the environment that directly affects hydrological, ecological, and climatic processes. However, owing to the influence of soil type, soil structure, topography, vegetation, and human activities, the distribution of soil water content is spatially heterogeneous. It is difficult to accurately monitor the distribution of soil moisture over large areas. To investigate the direct or indirect influence of various factors on soil moisture and obtain accurate soil moisture inversion results, we used structural equation models (SEMs) to determine the structural relationships between these factors and the degree of their influence on soil moisture. These models were subsequently transformed into the topology of artificial neural networks (ANN). Finally, a structural equation model coupled with an artificial neural network was constructed (SEM-ANN) for soil moisture inversion. The results showed the following: (1) The most important predictor of the spatial variability of soil moisture in the April was the temperature–vegetation dryness index, while land surface temperature was the most important predictor in the August; (2) After the ANN model was improved, the inversion accuracy of surface soil moisture by SEM-ANN model was improved, and the R2 of verification set was increased by 0.01 and 0.02 in April and August, respectively, and the relative analysis error was reduced by 0.5 % and 1.13 %. (3) There were no significant differences in soil moisture distribution trends between the April and August.

A Machine Learning-Based Method for Modelling a Proprietary SO2 Removal System in the Oil and Gas Sector

The aim of this study is to develop a model for a proprietary SO2 removal technology by using machine learning techniques and, more specifically, by exploiting the potentialities of artificial neural networks (ANNs). This technology is employed at the Eni oil and gas treatment plant in southern Italy. The amine circulating in this unit, that allows for a reduction in the SO2 concentration in the flue gases and to be compliant with the required specifications, is a proprietary solvent; thus, its composition is not publicly available. This has led to the idea of developing a machine learning (ML) algorithm for the unit description, with the objective of becoming independent from the licensor and more flexible in unit modelling. The model was developed in MatLab® by implementing ANNs and the aim was to predict three targets, namely the flow rate of SO2 that goes to the Claus unit, the emissions of SO2, and the flow rate of steam sent to the regenerator reboiler. These represent, respectively, the two physical outputs of the unit and a proxy variable of the amine quality. Three different models were developed, one for each target, that employed the Levenberg–Marquardt optimization algorithm. In addition, the ANN topology was optimized case by case. From the analysis of the results, it emerged that with a purely data-driven technique, the targets can be predicted with good accuracy. Therefore, this model can be employed to better manage the SO2 removal system, since it allows for the definition of an optimal control strategy and the maximization of the plant’s productivity by not exceeding the process constraints.

Particulate Matter Forecasting Using Different Deep Neural Network Topologies and Wavelets for Feature Augmentation

The concern about air pollution in urban areas has substantially increased worldwide. One of its main components, particulate matter (PM) with aerodynamic diameter of ≤2.5 µm (PM2.5), can be inhaled and deposited in deeper regions of the respiratory system, causing adverse effects on human health, which are even more harmful to children. In this sense, the use of deterministic and stochastic models has become a key tool for predicting atmospheric behavior and, thus, providing information for decision makers to adopt preventive actions to mitigate air pollution impacts. However, stochastic models present their own strengths and weaknesses. To overcome some of disadvantages of deterministic models, there has been an increasing interest in the use of deep learning, due to its simpler implementation and its success on multiple tasks, including time series and air quality forecasting. Thus, the objective of the present study is to develop and evaluate the use of four different topologies of deep artificial neural networks (DNNs), analyzing the impact of feature augmentation in the prediction of PM2.5 concentrations by using five levels of discrete wavelet transform (DWT). The following types of deep neural networks were trained and tested on data collected from two living lab stations next to high-traffic roads in Guildford, UK: multi-layer perceptron (MLP), long short-term memory (LSTM), one-dimensional convolutional neural network (1D-CNN) and a hybrid neural network composed of LSTM and 1D-CNN. The performance of each model in making predictions up to twenty-four hours ahead was quantitatively assessed through statistical metrics. The results show that wavelets improved the forecasting results and that discrete wavelet transform is a relevant tool to enhance the performance of DNN topologies, with special emphasis on the hybrid topology that achieved the best results among the applied models.