Neural Network Model Research Articles

Research on the Recognition of Piano-Playing Notes by a Music Transcription Algorithm

With the deepening research on music works, music transcription algorithms have been increasingly studied. This study examined the recognition of piano-playing notes using a music-transcription algorithm. First, the characteristics of MelSpec, LogSpec, and the constant Q-transform (CQT) are briefly introduced. Then, a convolutional recurrent neural network (CRNN) transcription algorithm, which includes four convolutional blocks and one bidirectional long short-term memory (BiLSTM) structure, was designed. The recognition performance of this method was analyzed using the MAPS dataset. LogSpec was found to have the best recognition performance for piano-playing notes when used as an input feature. In the CRNN structure, the recognition performance for piano-playing notes was the best when four convolutional blocks were used. Compared with the convolutional neural network (CNN), BiLSTM, and CNN-hidden Markov model algorithms, the F1-values of the CRNN algorithm were 84.9%, 92.24%, and 79.27% for frames, notes, and offsets, respectively, achieving the best recognition results. The results verify that the CRNN transcription algorithm is effective for the recognition of piano-playing notes and can be applied in practice.

Research on Cross-Border e-Commerce Supply Chain Prediction and Optimization Model Based on Convolutional Neural Network Algorithm

Enhancing the precision of supply chain management and reducing operational costs are crucial for the development of the cross-border e-commerce market. However, existing research often overlooks the demand uncertainty caused by seasonal variations and the challenges of handling returns in logistics. Therefore, this paper proposes a SARIMA-CNN-BiLSTM prediction model that effectively captures both the seasonal and nonlinear characteristics of cross-border e-commerce supply chains. Additionally, by incorporating the returns process, a supply chain distribution optimization model is developed with the objective of minimizing total operational costs. The model is solved using an improved whale optimization algorithm. In validation with real-world data, the SARIMA-CNN-BiLSTM model achieved a mean absolute percentage error reduction of 6.479 and 7.703 compared to convolutional neural network (CNN) and BiLSTM models, respectively. Moreover, the chosen optimization algorithm reduced the cost by 231,310 CNY, 62,564 CNY, and 131,632 CNY compared to the whale optimization algorithm, genetic algorithm, and particle swarm optimization, respectively. The proposed approach provides robust support for cross-border e-commerce enterprises in reducing costs and enhancing efficiency in their operations.

In-situ evaluation of hole quality and cutting tool condition in robotic drilling of composite materials using machine learning

Abstract The massive adoption of industrial robots in the manufacturing sector has significantly increased automation of installation and inspection procedures, particularly benefiting the aerospace industry, where large volumes of holes are drilled in each aircraft. However, mechanical drilling remains challenging when dealing with composite materials due to their inherently heterogeneous structure. This work presents a novel approach for in-situ hole quality inspection utilising integrated sensor data of an industrial robotic drill, combined with a machine learning model. Additionally, a novel classification approach for evaluating hole quality is proposed. This study employed a KUKA industrial robot, fitted with a multifunctional end-effector, to drill holes in a composite material used in aerospace applications. An ensemble neural network (ENN) model, which combines an artificial neural network with a genetic algorithm, was used to assess the quality of these drilled holes. The model was specifically developed and tested on the machined holes to relate process input parameters and drilling torques to hole quality. The model predictions were validated with six unseen datasets, of which five were predicted accurately. A full factorial study of the process parameters was conducted using analysis of variance (ANOVA) to investigate the relationship between tool condition and drilling torque. The results of the ANOVA show that tool condition is the largest contributor to drilling torque. The method proposed in this work, which allows real-time monitoring of hole quality, has the potential to improve manufacturing productivity of drilled components while ensuring high-quality end products.

A Comparative Study of Neural Network and Wavelet Decomposition Models for Price Forecasting of Tomato

Tomato price forecasting in Bihar marketplaces is the subject of a study. The study compares and contrasts many time series models. Three models are compared: WT-TDNN (Wavelet Transform with TDNN), TDNN (Time-Delay Neural Network), and ARIMA. This study evaluates the efficacy of advanced and traditional time series models for forecasting tomato prices in Bihar markets, with a focus on ARIMA, Time Delay Neural Networks (TDNN), and Wavelet-TDNN (WT-TDNN). ARIMA models were optimized using the Akaike Information Criterion (AIC), identifying (0, 1, 2) and (1, 1, 1) as optimal configurations, and validated through Wald's test. TDNN models with 2:4s:1l and 1:6s:1l architectures exhibited robust performance based on RMSE, MAE, and MAPE metrics. The WT-TDNN model, which integrates wavelet decomposition with TDNN, demonstrated superior predictive accuracy over both ARIMA and standard TDNN models. By employing the Haar filter for series decomposition into orthogonal components and applying TDNN to each component, WT-TDNN effectively captured the complex, non-linear dynamics of agricultural price series. This innovative approach enhances the precision of price forecasting, offering significant implications for agricultural economics and machine learning applications. The findings provide valuable insights for policymakers, farmers, and market analysts, enabling improved decision-making and fostering resilience in the agricultural sector. The study highlights WT-TDNN as a transformative tool for addressing the challenges of agricultural price prediction.

Assessment of Mango Canopy Water Content Through the Fusion of Multispectral Unmanned Aerial Vehicle (UAV) and Sentinel-2 Remote Sensing Data

This study proposes an Additive Wavelet Transform (AWT)-based method to fuse Multispectral UAV (MS UAV, 5 cm resolution) and Sentinel-2 satellite imagery (10–20 m resolution), generating 5 cm resolution fused images with a focus on near-infrared and shortwave infrared bands to enhance the accuracy of mango canopy water content monitoring. The fused Sentinel-2 and MS UAV data were validated and calibrated using field-collected hyperspectral data to construct vegetation indices, which were then used with five machine learning (ML) models to estimate Fuel Moisture Content (FMC), Equivalent Water Thickness (EWT), and canopy water content (CWC). The results indicate that the addition of fused Sentinel-2 data significantly improved the estimation accuracy of all parameters compared to using MS UAV data alone, with the Genetic Algorithm Backpropagation Neural Network (GABP) model performing best (R2 = 0.745, 0.859, and 0.702 for FMC, EWT, and CWC, respectively), achieving R2 improvements of 0.066, 0.179, and 0.210. Slope, canopy coverage, and human activities were identified as key factors influencing the spatial variability of FMC, EWT, and CWC, with CWC being the most sensitive to environmental changes, providing a reliable representation of mango canopy water status.

Utilizing a Novel Convolutional Neural Network for Diagnosis and Lesion Delineation in Colorectal Cancer Screening.

Early detection of colorectal cancer is vital for enhancing cure rates and alleviating treatment burdens. Nevertheless, the high demand for screenings coupled with a limited number of endoscopists underscores the necessity for advanced deep learning techniques to improve screening efficiency and accuracy. This study presents an innovative convolutional neural network (CNN) model, trained on 8260 images from screenings conducted at four medical institutions. The model incorporates parallel global and local feature extraction branches and a distinctive classification head, facilitating both cancer classification and the creation of heatmaps that outline cancerous lesion regions. Performance evaluations of the CNN model, measured against five leading models using accuracy, precision, recall, and F1 score, revealed its superior efficacy across these metrics. Furthermore, the heatmaps proved effective in aiding the automatic identification of lesion locations. In summary, this CNN model represents a promising advancement in early colorectal cancer screening, delivering precise, swift diagnostic results and robust interpretability through its automatic lesion highlighting capabilities.

GLAC-Unet: Global-Local Active Contour Loss with an Efficient U-Shaped Architecture for Multiclass Medical Image Segmentation.

The field of medical image segmentation powered by deep learning has recently received substantial attention, with a significant focus on developing novel architectures and designing effective loss functions. Traditional loss functions, such as Dice loss and Cross-Entropy loss, predominantly rely on global metrics to compare predictions with labels. However, these global measures often struggle to address challenges such as occlusion and nonuni-form intensity. To overcome these issues, in this study, we propose a novel loss function, termed Global-Local Active Contour (GLAC) loss, which integrates both global and local image features, reformulated within the Mumford-Shah framework and extended for multiclass segmentation. This approach enables the neural network model to be trained end-to-end while simultaneously segmenting multiple classes. In addition to this, we enhance the U-Net architecture by incorporating Dense Layers, Convolutional Block Attention Modules, and DropBlock. These improvements enable the model to more effectively combine contextual information across layers, capture richer semantic details, and mitigate overfitting, resulting in more precise segmentation outcomes. We validate our proposed method, namely GLAC-Unet, which utilizes the GLAC loss in conjunction with our modified U-shaped architecture, on three biomedical segmentation datasets that span a range of modalities, including two-dimensional and three-dimensional images, such as dermoscopy, cardiac magnetic resonance imaging, and brain magnetic resonance imaging. Extensive experiments demonstrate the promising performance of our approach, achieving a Dice score (DSC) of 0.9125 on the ISIC-2018 dataset, 0.9260 on the Automated Cardiac Diagnosis Challenge (ACDC) 2017, and 0.927 on the Infant Brain MRI Segmentation Challenge 2019. Furthermore, statistical significance testing with p-values consistently smaller than 0.05 on the ISIC-2018 and ACDC datasets confirms the superior performance of the proposed method compared to other state-of-the-art models. These results highlight the robustness and effectiveness of our multiclass segmentation technique, underscoring its potential for biomedical image analysis. Our code will be made available at https://github.com/minhnhattrinh312/Active-Contour-Loss-based-on-Global-and-Local-Intensity.

Biometric-Based Key Generation and User Authentication Using Voice Password Images and Neural Fuzzy Extractor

This work is devoted to the development of a biometric authentication system and the generation of a cryptographic key or a long password of 1024 bits based on a voice password, which ensures the protection of a biometric template from compromise. A new hybrid neural network model based on two types of trigonometric correlation neurons was proposed. The model is capable of recording correlation links between features and is resistant to data extraction attacks. The experiments were conducted on our own AIC-spkr-130 dataset and the publicly available RedDots, including recordings of user voices in different psycho-emotional states (sleepy state, alcohol intoxication). The results show that the proposed neural fuzzy extractor model provides an equal error probability level of EER = 2.1%.

Package Positioning Based on Point Registration Network DCDNet-Att

The application of robot technology in the automatic transportation process of packaging bags is becoming increasingly common. Point cloud registration is the key to applying industrial robots to automatic transportation systems. However, current point cloud registration models cannot effectively solve the registration of deformed targets like packaging bags. In this study, a new point cloud registration network, DCDNet-Att, is proposed, which uses a variable weight dynamic graph convolution module to extract point cloud features. A feature interaction module is used to extract common features between the source point cloud and the template point cloud. The same geometric features between the two pairs of point clouds are strengthened through a bottleneck module. A channel attention model is used to obtain the channel attention weights. The attention weight of each spatial position is calculated, and a rotation translation structure is used to sequentially obtain quaternions and translation vectors. A feature fitting loss function is used to constrain the parameters of the neural network model to have a larger receptive field. Compared with seven methods, including the ICP algorithm, GO-ICP algorithm, and FGR algorithm, the proposed method had rotation errors (MAE, RMSE, and Error of 1.458, 2.541, and 1.024 in the ModelNet40 dataset, respectively) and translation errors (MAE, RMSE, and Error of 0.0048, 0.0114, and 0.0174, respectively). When registering the ModelNet40 dataset with Gaussian noise, the rotation errors (MAE, RMSE, and Error) were 2.028, 3.437, and 2.478, respectively, and the translation errors (MAE, RMSE, and Error) were 0.0107, 0.0327, and 0.0285, respectively. The experimental results were superior to those of the other methods, and the model was effective at registering packaging bag point clouds.

An oral microbiota-based deep neural network model for risk stratification and prognosis prediction in gastric cancer

An oral microbiota-based deep neural network model for risk stratification and prognosis prediction in gastric cancer

Topographic Axes of Wiring Space Converge to Genetic Topography in Shaping Human Cortical Layout.

Genetic information is involved in the gradual emergence of cortical areas since the neural tube begins to form, shaping the heterogeneous functions of neural circuits in the human brain. Informed by invasive tract-tracing measurements, the cortex exhibits marked interareal variation in connectivity profiles, revealing the heterogeneity across cortical areas. However, it remains unclear about the organizing principles possibly shared by genetics and cortical wiring to manifest the spatial heterogeneity across cortex. Instead of considering a complex one-to-one mapping between genetic coding and interareal connectivity, we hypothesized the existence of a more efficient way that the organizing principles are embedded in genetic profiles to underpin the cortical wiring space. Leveraging vertex-wise tractography in diffusion-weighted MRI, we derived the global connectopies in both female and male to reliably index the organizing principles of interareal connectivity variation in a low-dimensional space, which captured three dominant topographic patterns along the dorsoventral, rostrocaudal, and mediolateral axes of the cortex. More importantly, we demonstrated that the global connectopies converge with the gradients of a vertex-by-vertex genetic correlation matrix on the phenotype of cortical morphology and the cortex-wide spatiomolecular gradients. By diving into the genetic profiles, we found that the critical role of genes scaffolding the global connectopies was related to brain morphogenesis and enriched in radial glial cells before birth and excitatory neurons after birth. Taken together, our findings demonstrated the existence of a genetically determined space that encodes the interareal connectivity variation, which may give new insights into the links between cortical connections and arealization.Significance Statement Genetic factors have involved the gradual emergence of cortical areas since the neural tube begins to form, shaping the specialization of neural circuitry in the human brain. However, the mechanisms through which genetics encode the complex interareal connectivity remain a pivotal and unanswered question in the field of neuroscience. Here, we hypothesized that a genetically determined space encoding the interareal connectivity variation exists, which may give new insights into the links between cortical connections and arealization. We combined diffusion tractography with a dimension reduction framework to unravel the underlying global topographic principle revealed by the anatomical connections.

Multidimensional free shape-morphing flexible neuromorphic devices with regulation at arbitrary points

Biological neural systems seamlessly integrate perception and action, a feat not efficiently replicated in current physically separated designs of neural-imitating electronics. This segregation hinders coordination and functionality within the neuromorphic system. Here, we present a flexible device tailored for neuromorphic computation and muscle actuation. Each individual device component emulates essential synaptic functions for neural computing, while the collective ensemble replicates muscle actuation in response to efferent neuromuscular commands. These properties stem from densely-packed, hydrophilic nanometer-sized channels, and the erection of a high-entropy, intricately silver nanowires to capture and store of hydrated cations. Leveraging the remarkable deformation effect, we demonstrate hazard detection-avoidance robot, and multidimensional integration for arbitrary programmed shapes like 360° panoramic information capture and soft-bodied biological deformations wherein localized responses to stimuli are harmoniously integrated to achieve arbitrary coordinated motion. These results provide a significant avenue for the development of future flexible electronics and bio-inspired systems.

Deep learning approach for optimizing material removal rate in wire electrical discharge machining using neural networks

Abstract This research focuses on optimizing the Material Removal Rate (MRR) in Wire Electrical Discharge Machining (Wire EDM) using deep learning techniques. The study identifies existing challenges in predicting MRR due to complex, non-linear relationships between process parameters. By adopting a systematic design of experiments, a dataset was generated to evaluate the impact of pulse on time, pulse off time, current, voltage, and wire feed rate on MRR. A neural network model was developed and trained with different activation functions, with the Sigmoid function yielding the best performance metrics, including an R-square of 0.9999 and a Mean Squared Error (MSE) of 0.0004. This work demonstrates the potential of artificial intelligence to enhance machining processes, providing valuable insights for industrial applications and paving the way for future research in advanced manufacturing technologies.

Predicting Chlorophyll-a Concentrations in the World's Largest Lakes Using Kolmogorov-Arnold Networks.

Accurate prediction of chlorophyll-a (Chl-a) concentrations, a key indicator of eutrophication, is essential for the sustainable management of lake ecosystems. This study evaluated the performance of Kolmogorov-Arnold Networks (KANs) along with three neural network models (MLP-NN, LSTM, and GRU) and three traditional machine learning tools (RF, SVR, and GPR) for predicting time-series Chl-a concentrations in large lakes. Monthly remote-sensed Chl-a data derived from Aqua-MODIS spanning September 2002 to April 2024 were used. The models were evaluated based on their forecasting capabilities from March 2024 to August 2024. KAN consistently outperformed others in both test and forecast (unseen data) phases and demonstrated superior accuracy in capturing trends, dynamic fluctuations, and peak Chl-a concentrations. Statistical evaluation using ranking metrics and critical difference diagrams confirmed KAN's robust performance across diverse study sites, further emphasizing its predictive power. Our findings suggest that the KAN, which leverages the KA representation theorem, offers improved handling of nonlinearity and long-term dependencies in time-series Chl-a data, outperforming neural network models grounded in the universal approximation theorem and traditional machine learning algorithms.

Applying convolutional neural networks for mustard variety recognition

The aim of this study was to develop and apply a Convolutional Neural Network (CNN) model to recognize and classify white mustard (Sinapis Alba L.) varieties, addressing the complex task of discriminating among 57 varieties. Utilizing a one-dimensional CNN model, the research focused on multivariate analysis based on a set of 15 traits. The CNN architecture included convolutional layers, batch normalization, pooling, flattening, dropout, and dense layers. The model demonstrated effectiveness in classifying varieties, achieving high accuracy and providing valuable insights into potential new varieties. Subset division, a new approach, was applied. Evaluation metrics, including accuracy, F1 score, precision, and recall, were calculated for eight subsets, confirming the model's robust performance. While this study uses mustard as an illustrative example, the method is not limited to this crop and can be extended to other agricultural crops, with potential modifications depending on the specific traits relevant to each crop. The approach contributes to agricultural advancements, offering a reliable tool for breeders to assess variety distinctness and streamline the testing process. The model’s ability to detect unknown varieties further enhances its utility in agricultural research covering a comprehensive and impactful advancement in variety classification.

Estimating Throughput in Optical Backbone Networks Using Deep Neural Networks

Optical backbone networks, which use optical fibres as the transmission medium, form the core infrastructure used by network operators to deliver services to users, as well as by Internet companies to route traffic between data centres. The network throughput is a key parameter in the analysis of the networks' performance. However, its determination can be a complex process that involves long computation times, since aspects related to both the physical and network layers need to be accounted for. To face this challenge, we propose a machine learning solution: a deep neural network (DNN) model, that has the goal of estimating the values of the network throughput and of a closely related parameter, average channel capacity, accurately and with short computation times. The simulation results indicate that the DNN model accurately predicts both outputs, with mean relative errors of 6.17% for the network throughput and 2.84% for the average channel capacity. These predictions are made in just a few milliseconds, providing a significant advantage over the heuristic routing algorithms, which can take up to tens of seconds in larger networks.

Peanut oil price change forecasts through the neural network

Purpose For a wide range of market actors, including policymakers, forecasting changes in commodity prices is crucial. As one of essential edible oil, peanut oil’s price swings are certainly important to predict. In this paper, the weekly wholesale price index for the period of January 1, 2010 to January 10, 2020 is used to address this specific forecasting challenge for the Chinese market. Design/methodology/approach The nonlinear auto-regressive neural network (NAR-NN) model is the forecasting method used. Forecasting performance based on various settings, such as training techniques, delay counts, hidden neuron counts and data segmentation ratios, are assessed to build the final specification. Findings With training, validation and testing root mean square errors of 5.89, 4.96 and 5.57, respectively, the final model produces reliable and accurate forecasts. Here, this paper demonstrates the applicability of the NAR-NN approach for commodity price predictions. Originality/value On the one hand, the findings may be used as independent technical price movement predictions. Conversely, they may be included in forecast combinations with forecasts derived from other models to form viewpoints of commodity price patterns for policy research.

Comparative analysis of deep learning models for crack detection in buildings

Life-time of the buildings is generally challenged by the act of nature. In-spite of the fact that the constructions provide minimum guarantee on quality and durability, certain mismatch in the composition of the materials, stress on the building, and chemical or physical imbalance of the materials, lead to surface crack. Cracks are also generated due to the shuffle of climatic conditions, which leads to the contraction and expansion of the building surfaces, and other damages. The guarantee on building safety and serviceability depends on how these buildings are successfully assessed and maintained. The development of Artificial Intelligence (AI) techniques, provide favourable solutions in-order to handle, manage and solve building cracks, through analysis using deep image neural network models, that perform classification of the building with crack images. As a result, a critical challenge for many civil engineering applications is the precise, quick, and automated identification of cracks on structural surfaces is addressed with the solutions provided by the deep image neural networks. In this research, we tackle the research gap and data scarcity by developing and curating a novel deep learning image processing for detecting cracks in brickwork. We also train and validate several deep learning models to classify brickwork images as either cracked or normal. The dataset of the proposed work contains 24,000 images which are classified through binary classes. These classes are generated for crack and non-crack images. The various parameters such as Batch size, Pooling, Activation functions Learning-rate, Kernel-Size, Normalization, and Optimizers are used for the evaluation of the model. The proposed work performs a comparative analysis of four deep image models such as Inception V3, VGG-16, RESNET-50 VGG-19, Inception ResNetV2 and CNN-RES MLP. With the analysis of all these models, the Inception V3 provides the best of all with the accuracy value of 99.98%. The InceptionV3 tops the Precision value of 99.99% and RESNET-50 tops the Recall value of 99.98%. The IncpetionV2 provided the best of the Region of Convergence value of 0.9999 which is the best among all the models for reliable and stable performance.

Predicting Paediatric Brain Disorders from MRI Images Using Advanced Deep Learning Techniques.

The problem at hand is the significant global health challenge posed by children's diseases, where timely and accurate diagnosis is crucial for effective treatment and management. Conventional diagnosis techniques are typical, use tedious processes and generate inaccurate results since they are executed by human beings and cause delays in treatment that can be fatal. Considering these and other shortcomings there exists a need to have more efficient and accurate solutions based on artificial intelligence. Machine learning and more specifically, deep learning algorithms are of great help in analysing medical and clinical images to detect as well as classify diseases. In this paper, we propose a system for detecting various childhood diseases using a range of advanced Convolutional Neural Network models like EfficientNetB0, EfficientNetB3, Xception, InceptionV3, MobileNetV2, VGG19, DenseNet169, ResNet50V2, ResNet152V2, and the hybrid architecture InceptionResNetV2. These models are trained on MRI images of paediatric brain disorders to achieve high prediction accuracy. We use data visualization techniques such as segmentation and contour-based feature extraction to extract regions of interest before feeding the data into the models. The models are optimized using both ADAM and RMSprop optimizers. EfficientNetB0, when optimized with RMSprop, achieves an accuracy of 94.59%, a loss of 0.44, and an RMSE of 0.66. InceptionResNetV2, optimized with ADAM, achieves the highest accuracy of 97.59%, while EfficientNetB0 demonstrates the lowest loss (0.25) and RMSE (0.5). We also evaluate the models based on their precision, learning curves, recall, computational time, and F1 score, highlighting the effectiveness of AI-driven approaches for the diagnosis and management of children's diseases.

Robust estimation of skin physiological parameters from hyperspectral images using Bayesian neural networks.

Machine learning models for the direct extraction of tissue parameters from hyperspectral images have been extensively researched recently, as they represent a faster alternative to the well-known iterative methods such as inverse Monte Carlo and inverse adding-doubling (IAD). We aim to develop a Bayesian neural network model for robust prediction of physiological parameters from hyperspectral images. We propose a two-component system for extracting physiological parameters from hyperspectral images. First, our system models the relationship between the measured spectra and the tissue parameters as a distribution rather than a point estimate and is thus able to generate multiple possible solutions. Second, the proposed tissue parameters are then refined using the neural network that approximates the biological tissue model. The proposed model was tested on simulated and in vivo data. It outperformed current models with an overall mean absolute error of 0.0141 and can be used as a faster alternative to the IAD algorithm. Results suggest that Bayesian neural networks coupled with the approximation of a biological tissue model can be used to reliably and accurately extract tissue properties from hyperspectral images on the fly.