Convolutional Filters Research Articles

Versatile time-window sliding machine learning techniques for stock market forecasting

The stock market which is a critical instrument in the modern financial system consistently attracts a significant number of individuals and financial institutions. The amalgamation of machine learning with stock forecasting has seen increased interest due to the rising prominence of machine learning. Among numerous machine learning models, the long short-term memory (LSTM) is favored by many researchers for its superior performance in long time series. This paper presents an innovative approach that integrates LSTM with versatile sliding window techniques to enhance prediction results and training performance. Moreover, the attached experiments incorporate convolutional filters and combined bivariate performance measures, which are invaluable methodologies for enhancing stock forecasting systems. These significant contributions are expected to be beneficial for researchers operating within this domain.

UAV imaging hyperspectral for barnyard identification and spatial distribution in paddy fields

Barnyard grass (Echinochloa crus-galli), a noxious weed prevalent in rice fields, poses a significant threat to rice growth and development, ultimately leading to substantial yield reductions. Its morphological resemblance to rice plants complicates its identification and management. Consequently, accurately identifying and mapping barnyard grass in intricate rice field environments holds substantial research value and significance. Furthermore, most existing weed detection studies rely on single-day data, limiting the models’ generalizability. Therefore, incorporating multi-temporal data into modeling is highly desirable. This study utilized an unmanned aerial vehicle (UAV) to capture multi-temporal, low-altitude hyperspectral imagery of paddy fields, which were subsequently stitched, calibrated, and filtered using the SG convolution filter. The continuous projection algorithm (SPA) was employed to extract sensitive bands for discriminating between rice and barnyard grass. Initially, single-time data was utilized for barnyard grass identification, with models such as support vector machine (SVM), random forest (RF), 1D convolutional neural network (1DCNN), and 3D convolutional neural network (3DCNN) constructed using both full-spectrum and feature bands. The results indicated that the SPA-3DCNN model exhibited the highest performance in identifying rice (F1-score: 0.942) and barnyard grass (F1-score: 0.8936). Seven selected feature bands (482.5234, 546.5415, 675.0806, 709.1382, 762.0431, 922.0157, and 944.6371 nm) were found to be highly discriminative for distinguishing barnyard grass from rice. Subsequently, multi-temporal data modeling was employed to enhance the models’ generalization ability, resulting in improved overall accuracy (OA) and Kappa coefficients for each model, along with reduced misclassification rates. The spatial distribution maps also exhibited better performance with increased temporal data. The infestation trend of barnyard grass was visualized using multi-temporal difference maps, and a density map of barnyard grass was generated through image binarization. This study successfully demonstrated the feasibility of utilizing UAV-borne hyperspectral imagery for identifying barnyard grass in complex paddy field environments. By improving the model’s generalization ability through multi-temporal modeling, the study provided robust data support for managing and preventing barnyard grass infestations, including the creation of spatial distribution and density maps for barnyard grass.

Graph Feature Refinement and Fusion in Transformer for Structural Damage Detection.

Structural damage detection is of significance for maintaining the structural health. Currently, data-driven deep learning approaches have emerged as a highly promising research field. However, little progress has been made in studying the relationship between the global and local information of structural response data. In this paper, we have presented an innovative Convolutional Enhancement and Graph Features Fusion in Transformer (CGsformer) network for structural damage detection. The proposed CGsformer network introduces an innovative approach for hierarchical learning from global to local information to extract acceleration response signal features for structural damage representation. The key advantage of this network is the integration of a graph convolutional network in the learning process, which enables the construction of a graph structure for global features. By incorporating node learning, the graph convolutional network filters out noise in the global features, thereby facilitating the extraction to more effective local features. In the verification based on the experimental data of four-story steel frame model experiment data and IASC-ASCE benchmark structure simulated data, the CGsformer network achieved damage identification accuracies of 92.44% and 96.71%, respectively. It surpassed the existing traditional damage detection methods based on deep learning. Notably, the model demonstrates good robustness under noisy conditions.

A detailed analysis of the interpretability of Convolutional Neural Networks for text classification

Abstract Convolutional Neural Networks (CNNs) have become ubiquitous in many NLP tasks. However, understanding its process is still an area with much to be done. In this paper, we introduce a method to study the interpretability of CNNs when used for text classification. More specifically, we work on the interpretability of the convolutional filters in the context of sentiment analysis. The framework used in this paper has allowed the understanding of the mechanics of the network when applied to the task of sentiment analysis. The results of the experiments reveal how certain parts of speech (POS) tags are more relevant than others to the classification of a sentence. Furthermore, we also observed a preference for shorter $n$-grams when classifying negative sentiment sequences. Additionally, we detected a certain amount of redundancy among the convolutional filters, thus allowing us to conclude that smaller architectures would have worked for this particular task. This was achieved by computing the relevant metrics to measure the influence of the pertinent property on the desired class. In our case, the concepts treated were POS tags that carry semantic information and properties were related to the convolutional filters.

A spatial-varying non-isotropic Gaussian-based convolution filter for smoothing GRACE-like temporal gravity fields

A spatial-varying non-isotropic Gaussian-based convolution filter for smoothing GRACE-like temporal gravity fields

Conditional computation in neural networks: Principles and research trends

This article summarizes principles and ideas from the emerging area of applying conditional computation methods to the design of neural networks. In particular, we focus on neural networks that can dynamically activate or de-activate parts of their computational graph conditionally on their input. Examples include the dynamic selection of, e.g., input tokens, layers (or sets of layers), and sub-modules inside each layer (e.g., channels in a convolutional filter). We first provide a general formalism to describe these techniques in an uniform way. Then, we introduce three notable implementations of these principles: mixture-of-experts (MoEs) networks, token selection mechanisms, and early-exit neural networks. The paper aims to provide a tutorial-like introduction to this growing field. To this end, we analyze the benefits of these modular designs in terms of efficiency, explainability, and transfer learning, with a focus on emerging applicative areas ranging from automated scientific discovery to semantic communication.

Non-intrusive method for audio quality assessment of lossy-compressed music recordings using convolutional neural networks

Most of the existing algorithms for the objective audio quality assessment are intrusive, as they require access both to an unimpaired reference recording and an evaluated signal. This feature excludes them from many practical applications. In this paper, we introduce a non-intrusive audio quality assessment method. The proposed method is intended to account for audio artefacts arising from the lossy compression of music signals. During its development, 250 high-quality uncompressed music recordings were collated. They were subsequently processed using the selection of five popular audio codecs, resulting in the repository of 13,000 audio excerpts representing various levels of audio quality. The proposed non-intrusive method was trained with the data obtained employing a well-established intrusive model (ViSQOL v3). Next, the performance of the trained model was evaluated utilizing the quality scores obtained in the subjective listening tests undertaken remotely over the Internet. The listening tests were carried out in compliance with the MUSHRA recommendation (ITU-R BS.1534-3). In this study, the following three convolutional neural networks were compared: (1) a model employing 1D convolutional filters, (2) an Inception-based model, and (3) a VGG-based model. The last-mentioned model outperformed the model employing 1D convolutional filters in terms of predicting the scores from the listening tests, reaching a correlation value of 0.893. The performance of the Inceptionbased model was similar to that of the VGG-based model. Moreover, the VGG-based model outperformed the method employing a stacked gated-recurrent-unit-based deep learning framework, recently introduced by Mumtaz et al. (2022).

Lung Disease Detection Using U-Net Feature Extractor Cascaded by Graph Convolutional Network.

Computed tomography (CT) scans have recently emerged as a major technique for the fast diagnosis of lung diseases via image classification techniques. In this study, we propose a method for the diagnosis of COVID-19 disease with improved accuracy by utilizing graph convolutional networks (GCN) at various layer formations and kernel sizes to extract features from CT scan images. We apply a U-Net model to aid in segmentation and feature extraction. In contrast with previous research retrieving deep features from convolutional filters and pooling layers, which fail to fully consider the spatial connectivity of the nodes, we employ GCNs for classification and prediction to capture spatial connectivity patterns, which provides a significant association benefit. We handle the extracted deep features to form an adjacency matrix that contains a graph structure and pass it to a GCN along with the original image graph and the largest kernel graph. We combine these graphs to form one block of the graph input and then pass it through a GCN with an additional dropout layer to avoid overfitting. Our findings show that the suggested framework, called the feature-extracted graph convolutional network (FGCN), performs better in identifying lung diseases compared to recently proposed deep learning architectures that are not based on graph representations. The proposed model also outperforms a variety of transfer learning models commonly used for medical diagnosis tasks, highlighting the abstraction potential of the graph representation over traditional methods.

The Detection of Railheads: An Innovative Direct Image Processing Method

This study presents a fully automated railhead detection method based on a direct image processing algorithm for use on a railway track. This method functions at a much faster pace than artificial intelligence algorithms that process rail images on embedded systems or low-power devices, as it does not require the use of significant computing resources. With the use of this method, railheads can be analyzed to identify the presence of cracks and other defects. We converted color images to halftone images, performed histogram equalizations to improve the contrast, applied a Gaussian filter to reduce the presence of noise, utilized convolutional filters to extract any vertical and horizontal lines, applied the Canny method and Sobel filters to refine the boundaries of the extracted lines, applied the Hough transform technique to extract lines belonging to the railhead images, and identified the segments with the highest brightness values to process the images of the railheads under study. The method of railhead separation described in this article will allow for further comprehensive diagnostics of the condition of rail threads to ensure the safe and sustainable operation of railway transport. The implementation of intelligent maintenance systems and effective monitoring of railway track conditions can reduce the negative impact on the environment and contribute to the advancement of rail transport as a sustainable, safe, and more environmentally friendly mode of transportation.

Marine oil spill detection and segmentation in SAR data with two steps Deep Learning framework

Marine oil spills pose significant ecological and economic threats worldwide, requiring effective decision-making tools. In this study, the optimal parameters, and configurations for Deep Learning models in oil spill classification and segmentation using Sentinel-1 SAR imagery were identified. First, a new Sentinel-1 image dataset was created. Ninety CNN configurations were explored for classification by varying the number of convolutional layers, filters, hidden layers, and neurons in each layer. For segmentation tasks, MLP and U-Net models were evaluated with variations in convolutional layers, filters, and incorporation of IoU and Focal Loss. The results indicated that a CNN model with six layers, 32 filters, and two hidden layers achieved 99 % classification accuracy. For segmentation, the U-Net model with more layers and filters using Focal Loss achieved 99 % accuracy and 96 % IoU. Therefore, a CNN and U-Net framework was proposed that achieves an overall accuracy of 95 % and an IoU of 90 %.

KepSalinst: Using Peripheral Points to Delineate Salient Instances.

Salient instance segmentation (SIS) is an emerging field that evolves from salient object detection (SOD), aiming at identifying individual salient instances using segmentation maps. Inspired by the success of dynamic convolutions in segmentation tasks, this article introduces a keypoints-based SIS network (KepSalinst). It employs multiple keypoints, that is, the center and several peripheral points of an instance, as effective geometrical guidance for dynamic convolutions. The features at peripheral points can help roughly delineate the spatial extent of the instance and complement the information inside the central features. To fully exploit the complementary components within these features, we design a differentiated patterns fusion (DPF) module. This ensures that the resulting dynamic convolutional filters formed by these features are sufficiently comprehensive for precise segmentation. Furthermore, we introduce a high-level semantic guided saliency (HSGS) module. This module enhances the perception of saliency by predicting a map for the input image to estimate a saliency score for each segmented instance. On four SIS datasets (ILSO, SOC, SIS10K, and COME15K), our KepSalinst outperforms all previous models qualitatively and quantitatively.

Solution to the cocktail party problem: A time-reversal active metasurface for multipoint focusing

The cocktail party effect is the capability to focus one’s auditory attention on particular audio sources while ignoring other audio sources. We propose an experimental strategy to reproduce this ability by designing a time-dependent metasurface composed of independent active mirrors. Each active mirror is a programmable acoustic unit cell capable of hearing, computing, and re-emitting acoustic signals: each of them acts as a convolution filter. The proper configuration of the metasurface temporal filters allows one to establish an acoustic communication link between groups of individuals immersed in the noisy environment: a multiple-user multiple-input, multiple-output acoustic system is built. Published by the American Physical Society 2024

DeepGCN based on variable multi‐graph and multimodal data for ASD diagnosis

AbstractDiagnosing individuals with autism spectrum disorder (ASD) accurately faces great challenges in clinical practice, primarily due to the data's high heterogeneity and limited sample size. To tackle this issue, the authors constructed a deep graph convolutional network (GCN) based on variable multi‐graph and multimodal data (VMM‐DGCN) for ASD diagnosis. Firstly, the functional connectivity matrix was constructed to extract primary features. Then, the authors constructed a variable multi‐graph construction strategy to capture the multi‐scale feature representations of each subject by utilising convolutional filters with varying kernel sizes. Furthermore, the authors brought the non‐imaging information into the feature representation at each scale and constructed multiple population graphs based on multimodal data by fully considering the correlation between subjects. After extracting the deeper features of population graphs using the deep GCN(DeepGCN), the authors fused the node features of multiple subgraphs to perform node classification tasks for typical control and ASD patients. The proposed algorithm was evaluated on the Autism Brain Imaging Data Exchange I (ABIDE I) dataset, achieving an accuracy of 91.62% and an area under the curve value of 95.74%. These results demonstrated its outstanding performance compared to other ASD diagnostic algorithms.

Composite convolution: A flexible operator for deep learning on 3D point clouds

Deep neural networks require specific layers to process point clouds, as the scattered and irregular location of 3D points prevents the use of conventional convolutional filters. We introduce the composite layer, a flexible and general alternative to the existing convolutional operators that process 3D point clouds. We design our composite layer to extract and compress the spatial information from the 3D coordinates of points and then combine this with the feature vectors. Compared to mainstream point-convolutional layers such as ConvPoint and KPConv, our composite layer guarantees greater flexibility in network design and provides an additional form of regularization. To demonstrate the generality of our composite layers, we define both a convolutional composite layer and an aggregate version that combines spatial information and features in a nonlinear manner, and we use these layers to implement CompositeNets. Our experiments on synthetic and real-world datasets show that, in both classification, segmentation, and anomaly detection, our CompositeNets outperform ConvPoint, which uses the same sequential architecture, and achieve similar results as KPConv, which has a deeper, residual architecture. Moreover, our CompositeNets achieve state-of-the-art performance in anomaly detection on point clouds. Our code is publicly available at https://github.com/sirolf-otrebla/CompositeNet.

On Minimizers and Convolutional Filters: Theoretical Connections and Applications to Genome Analysis.

Minimizers and convolutional neural networks (CNNs) are two quite distinct popular techniques that have both been employed to analyze categorical biological sequences. At face value, the methods seem entirely dissimilar. Minimizers use min-wise hashing on a rolling window to extract a single important k-mer feature per window. CNNs start with a wide array of randomly initialized convolutional filters, paired with a pooling operation, and then multiple additional neural layers to learn both the filters themselves and how they can be used to classify the sequence. In this study, our main result is a careful mathematical analysis of hash function properties showing that for sequences over a categorical alphabet, random Gaussian initialization of convolutional filters with max-pooling is equivalent to choosing a minimizer ordering such that selected k-mers are (in Hamming distance) far from the k-mers within the sequence but close to other minimizers. In empirical experiments, we find that this property manifests as decreased density in repetitive regions, both in simulation and on real human telomeres. We additionally train from scratch a CNN embedding of synthetic short-reads from the SARS-CoV-2 genome into 3D Euclidean space that locally recapitulates the linear sequence distance of the read origins, a modest step toward building a deep learning assembler, although it is at present too slow to be practical. In total, this article provides a partial explanation for the effectiveness of CNNs in categorical sequence analysis.

An Improved Multi-Scale Region of Interest Deep Learning Model for Fruit Disease Classification

One of the main sources of nutrients is fruits, yet a variety of diseases appear that harmfully impact the fruit's quality and productivity. To avoid this issue, Deep Learning (DL) models have been developed, which identify and classify fruit diseases earlier. Amongst, a new Attribute registration and Shadow removal Adversarial Network with Multi-scale Region-Of-Interest (ROI)-based Deep Learning (ArSAN-MRDL) model creates more shadow-less fruit images to classify multiple diseases in a single fruit image and localizes the diseased regions precisely. In contrast, the standard convolution filters in the MRDL may lose the discrimination information for similar fruit diseases, which degrades the classification accuracy. Hence, this article designs the ArSAN with an improved MRDL (ArSAN-iMRDL) model to improve the accuracy of classifying similar diseases in the single shadow-less fruit image. Important advances made in this paper include (i) isolating the channel and spatial interdependency prediction of downsampling to prevent information loss, (ii) replacing the convolution in MRDL's Residual Blocks (ResBlocks) with a pyramid convolution with dilated convolution at many dilation rates to enrich the finite-size representation, and (iii) replacing the convolution in MRDL's Residual Blocks (ResBlocks. This iMRDL increases the robustness of classifying and localizing multiple fruit diseases. At last, extensive experiments show that the ArSAN-iMRDL model on apple, citrus, and tomato fruit images reaches 95.15%, 95.6%, and 95.43% accuracy, correspondingly, compared to the existing CNN models.

Soft independence guided filter pruning

Filter pruning (FP) is an effective method for reducing the computational costs of convolutional neural networks, and herein, the most critical task involves evaluating the significance of each convolutional filter and eliminating the less important ones while minimizing performance degradation. Most existing FP methods consider only local information, which may prevent them from accurately recognizing the most important filters. To address this limitation, we propose the soft filter independence (SFI) method, which leverages global information to identify the most important filters using their magnitude and correlation information in different functional layers. The SFI criterion measures the replaceability of filters from a global perspective in a network. Filters with low independence can be represented effectively by others, so their information can be accurately conveyed by other filters. In addition, we introduce a novel SFI-based asymptotic pruning ratio, which improves training and pruning stability. Compared to the most advanced FP methods, our method enables CNNs to achieve higher pruning rates and better classification performance.

A GNN-based fraud detector with dual resistance to graph disassortativity and imbalance

With the prosperity of Internet services, various fraudulent activities have emerged, and some graph neural network-based methods have been proposed for fraud detection. These methods have achieved significant performances based on the assortativity assumption where the connected nodes tend to have the same labels. However, the fraud graphs are not always assortative but more likely disassortative as fraudsters usually disguise themselves by deliberately making extensive connections to benign users. Furthermore, some studies have noticed node imbalance issues as the fraudsters are far fewer than normal users. But for another one, the edge imbalance issue is unexplored, and they also failed to make full use of these rare but valuable edges. Therefore, we propose an imbalanced disassortative graph learning framework for fraud detection. Specifically, a learnable dual-channel graph convolution filter is introduced to adaptively aggregate low- and high-frequency signals from its neighbors to assimilate/discriminate nodes with assortative/disassortative edges. Then, the label-aware node and edge samplers are designed to remedy graph imbalance problems. Furthermore, the sampled edges are treated as the auxiliary supervision signal to explicitly facilitate the training of graph filters. Extensive experiments on real-world datasets verify the superiority of our method.

Design of intelligent assistive system for physical education: based on personalized training plan

Instead of viewing different components in solitude, teachers' opinion with respect and pedagogical knowledge demands a comprehension of the relationships connecting them. The instructor must be knowledgeable about the subject matter, comprehend the most effective delivery methods for the information, be acquainted with the traits of the children, and be aware of the academic setting to educate in that field effectively. Sport is still undergoing economic, cultural, and ethical change. On the other side, throughout the past century, science has been the sport's most pervasive transformation. Players can now jump higher, run faster, and, importantly—retain their health due to scientific understanding. Even though scholars, organizations, and governments have indeed encouraged physical education instructors to incorporate software in their classes, technology is often employed for routine duties, such as tracking enrollment and assessing, recording, and reporting children’s' progress. In order to estimate posture in physical training, this research proposes a continuous filter convolution neural network. The approach also assesses the learners' understanding, memory, and accomplishments and offers suggestions for enhancements and remedial actions. The framework and conventional teaching-learning methodologies are then compared characteristics per aspect for output criteria. Finally, the classification algorithm is contrasted with other deep learning algorithms, and it is found that the proposed ConFil_CNN achieves 98.5% accuracy, 96.7% of precision, 93.5% of recall, 95.2% of F1-score and 12.4ms of response time.

Convolutional Spectral Kernel Learning with Generalization Guarantees (Abstract Reprint)

Kernel methods are powerful tools to capture nonlinear patterns behind given data but often lead to poor performance on complicated tasks compared to convolutional neural networks. The reason is that kernel methods are still shallow and fully connected models, failing to reveal hierarchical features and local interdependencies. In this paper, to acquire hierarchical and local knowledge, we incorporate kernel methods with deep architectures and convolutional operators in a spectral kernel learning framework. Based on the inverse Fourier transform and Rademacher complexity theory, we provide the generalization error bounds for the proposed model and prove that under suitable initialization, deeper networks lead to tighter error bounds. Inspired by theoretical findings, we finally completed the convolutional spectral kernel network (CSKN) with two additional regularizers and an initialization strategy. Extensive ablation results validate the effectiveness of non-stationary spectral kernel, multiple layers, additional regularizers, and the convolutional filters, which coincide with our theoretical findings. We further devise a VGG-type 8-layers CSKN, and it outperforms the existing kernel-based networks and popular CNN models on the medium-sized image classification tasks.