Convolutional Neural Network Filter Research Articles

Malayalam Handwritten Words Recognition: A Review

This review examines character segmentation and offers an elegant method for identifying and transforming handwritten Malayalam words from picture documents into text. Character touchings, different writing styles, and noisy, damaged scanned photos make it difficult to recognise handwritten text. Taking use of today's world of rich data and algorithmic developments, the system uses deep convolutional neural networks (CNNs) to address these challenges. The three steps of Malayalam handwritten word recognition are segmentation, recognition, and pre-processing. Making Malayalam character datasets is the first stage, and then pre-processing to improve image quality comes next. Then, in order to maximise the system's capacity to precisely forecast Malayalam characters, a CNN model is built to extract relevant information. The last phase of the recognition process involves the system classifying the characters. This project is significant since it uses CNN filters to enhance feature recognition, which enhances the accuracy of Malayalam character prediction. Key Words: Deep Learning, Deep Convolution Neural Network (DCNN), Character recognition, Character segmentation,

Performance analysis of hybrid deep learning framework using a vision transformer and convolutional neural network for handwritten digit recognition

Digitization created a demand for highly efficient handwritten document recognition systems. A handwritten document consists of digits, text, symbols, diagrams, etc. Digits are an essential element of handwritten documents. Accurate recognition of handwritten digits is vital for effective communication and data analysis. Various researchers have attempted to address this issue with modern convolutional neural network (CNN) techniques. Even after training, CNN filter weights remain unchanged despite the high identification accuracy. As a result, the process cannot flexibly adapt to input changes. Hence computer vision researchers have recently become interested in Vision Transformers (ViTs) and Multilayer Perceptrons (MLPs). The shortcomings of CNNs gave rise to a hybrid model revolution that combines the best elements of the two fields. This paper analyzes how the hybrid convolutional ViT model affects the ability to recognize handwritten digits. Also, the real-time data contains noise, distortions, and varying writing styles. Hence, cleaned and uncleaned handwritten digit images are used for evaluation in this paper. The accuracy of the proposed method is compared with the state-of-the-art techniques, and the result shows that the proposed model achieves the highest recognition accuracy. Also, the probable solutions for recognizing other aspects of handwritten documents are discussed in this paper.•Analyzed the effect of convolutional vision transformer on cleaned and real-time handwritten digit images.•The model's performance improved with the implication of cross-validation and hyper-parameter tuning.•The results show that the proposed model is robust, feasible, and effective on cleaned and uncleaned handwritten digits.

Automated Detection of Cerebral Microbleeds on Two-dimensional Gradient-recalled Echo T2* Weighted Images Using a Morphology Filter Bank and Convolutional Neural Network.

We present a novel algorithm for the automated detection of cerebral microbleeds (CMBs) on 2D gradient-recalled echo T2* weighted images (T2*WIs). This approach combines a morphology filter bank with a convolutional neural network (CNN) to improve the efficiency of CMB detection. A technical evaluation was performed to ascertain the algorithm's accuracy. In this retrospective study, 60 patients with CMBs on T2*WIs were included. The gold standard was set by three neuroradiologists based on the Microbleed Anatomic Rating Scale guidelines. Images with CMBs were extracted from the training dataset comprising 30 cases using a morphology filter bank, and false positives (FPs) were removed based on the threshold of size and signal intensity. The extracted images were used to train the CNN (Vgg16). To determine the effectiveness of the morphology filter bank, the outcomes of the following two methods for detecting CMBs from the 30-case test dataset were compared: (a) employing the morphology filter bank and additional FP removal and (b) comprehensive detection without filters. The trained CNN processed both sets of initial CMB candidates, and the final CMB candidates were compared with the gold standard. The sensitivity and FPs per patient of both methods were compared. After CNN processing, the morphology-filter-bank-based method had a 95.0% sensitivity with 4.37 FPs per patient. In contrast, the comprehensive method had a 97.5% sensitivity with 25.87 FPs per patient. Through effective CMB candidate refinement with a morphology filter bank and FP removal with a CNN, we achieved a high CMB detection rate and low FP count. Combining a CNN and morphology filter bank may facilitate the accurate automated detection of CMBs on T2*WIs.

IIRI-Net: An interpretable convolutional front-end inspired by IIR filters for speaker identification

Learning interpretable filters in Convolutional Neural Networks (CNNs) is an approach that helps to build models with better generalization ability. Interpretable filters can reveal some hidden aspects of the task and help to improve the model. One of the most successful approaches in the field of the speech processing is SincNet, where the model learns some band-pass filters in the first layer of a CNN with a raw waveform as its input. In this paper, similar to SincNet, some meaningful filters are proposed, which here are inspired by Infinite Impulse Response (IIR) filters. The proposed model uses a phase correction process to ensure that phase linearity is satisfied. The effective length of the truncated IIR filter is calculated based on the accumulated energy, and the effect of changing the filter size on the final results has been investigated. The proposed model is evaluated in the speaker identification task on the TIMIT and Librispeech datasets and compared with traditional CNNs and four interpretable kernel-based models. The experimental results show the superiority of the proposed model both in performance and convergence speed. Moreover, some patterns of the speech signal, which lead to uniquely identifying a speaker, are analyzed by examining the spectrum of the learned filters.

Enhancement of Product-Inspection Accuracy Using Convolutional Neural Network and Laplacian Filter to Automate Industrial Manufacturing Processes

The automation of the manufacturing process of printed circuit boards (PCBs) requires accurate PCB inspections, which in turn require clear images that accurately represent the product PCBs. However, if low-quality images are captured during the involved image-capturing process, accurate PCB inspections cannot be guaranteed. Therefore, this study proposes a method to effectively detect defective images for PCB inspection. This method involves using a convolutional neural network (CNN) and a Laplacian filter to achieve a higher accuracy of the classification of the obtained images as normal and defective images than that obtained using existing methods, with the results showing an improvement of 11.87%. Notably, the classification accuracy obtained using both a CNN and Laplacian filter is higher than that obtained using only CNNs. Furthermore, applying the proposed method to images of computer components other than PCBs results in a 5.2% increase in classification accuracy compared with only using CNNs.

A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures.

Some recent studies show that filters in convolutional neural networks (CNNs) have low color selectivity in datasets of natural scenes such as Imagenet. CNNs, bio-inspired by the visual cortex, are characterized by their hierarchical learning structure which appears to gradually transform the representation space. Inspired by the direct connection between the LGN and V4, which allows V4 to handle low-level information closer to the trichromatic input in addition to processed information that comes from V2/V3, we propose the addition of a long skip connection (LSC) between the first and last blocks of the feature extraction stage to allow deeper parts of the network to receive information from shallower layers. This type of connection improves classification accuracy by combining simple-visual and complex-abstract features to create more color-selective ones. We have applied this strategy to classic CNN architectures and quantitatively and qualitatively analyzed the improvement in accuracy while focusing on color selectivity. The results show that, in general, skip connections improve accuracy, but LSC improves it even more and enhances the color selectivity of the original CNN architectures. As a side result, we propose a new color representation procedure for organizing and filtering feature maps, making their visualization more manageable for qualitative color selectivity analysis.

Multivariate Time Series Forecasting with Transfer Entropy Graph

Multivariate Time Series (MTS) forecasting is an essential problem in many fields. Accurate forecasting results can effectively help in making decisions. To date, many MTS forecasting methods have been proposed and widely applied. However, these methods assume that the predicted value of a single variable is affected by all other variables, ignoring the causal relationship among variables. To address the above issue, we propose a novel end-to-end deep learning model, termed graph neural network with neural Granger causality, namely CauGNN, in this paper. To characterize the causal information among variables, we introduce the neural Granger causality graph in our model. Each variable is regarded as a graph node, and each edge represents the casual relationship between variables. In addition, convolutional neural network filters with different perception scales are used for time series feature extraction, to generate the feature of each node. Finally, the graph neural network is adopted to tackle the forecasting problem of the graph structure generated by the MTS. Three benchmark datasets from the real world are used to evaluate the proposed CauGNN, and comprehensive experiments show that the proposed method achieves state-of-the-art results in the MTS forecasting task.

FP-AGL: Filter Pruning With Adaptive Gradient Learning for Accelerating Deep Convolutional Neural Networks

Filter pruning is a technique that reduces computational complexity, inference time, and memory footprint by removing unnecessary filters in convolutional neural networks (CNNs) with an acceptable drop in accuracy, consequently accelerating the network. Unlike traditional filter pruning methods utilizing zeroing-out filters, we propose two techniques to achieve the effect of pruning more filters with less performance degradation, inspired by the existing research on centripetal stochastic gradient descent (C-SGD), wherein the filters are removed only when the ones that need to be pruned have the same value. First, to minimize the negative effect of centripetal vectors that gradually make filters come closer to each other, we redesign the vectors by considering the effect of each vector on the loss-function using the Taylor-based method. Second, we propose an adaptive gradient learning (AGL) technique that updates weights while adaptively changing the gradients. Through AGL, performance degradation can be mitigated because some gradients maintain their original direction, and AGL also minimizes the accuracy loss by perfectly converging the filters, which require pruning, to a single point. Finally, we demonstrate the superiority of the proposed method on various datasets and networks. In particular, on the ILSVRC-2012 dataset, our method removed 52.09% FLOPs with a negligible 0.15% top-1 accuracy drop on ResNet-50. As a result, we achieve the most outstanding performance compared to those reported in previous studies in terms of the trade-off between accuracy and computational complexity.

NNFacet: Splitting Neural Network for Concurrent Smart Sensors

Various deep neural networks (DNNs) including convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have shown appealing performance in various classification tasks. However, due to their large sizes, a single DNN often cannot fit into the memory of resource-constrained smart IoT sensors. This paper presents a DNN splitting framework called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">NNFacet</i> that aims to run a DNN-based classification task on a total of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$N$</tex-math></inline-formula> concurrent battery-based sensors observing the same physical process. We begin with determining the importance of all CNN filters or RNN units in learning each class. Then, an optimization problem divides the class set into <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$N$</tex-math></inline-formula> subsets and assigns them to the sensors, where the important CNN filters or RNN units associated with a class subset form a small model that is deployed to a sensor. Lastly, a multilayer perceptron is trained and deployed to a cloud or edge server, which yields the final classification result based on the low-dimensional features extracted by the sensors using their small models for the same observation. We apply NNFacet to three case studies of voice sensing, vibration sensing, and visual sensing. Extensive evaluation shows that NNFacet outperforms four baseline approaches in terms of system lifetime, latency, and classification accuracy.

Phase Compensation for Continuous Variable Quantum Key Distribution Based on Convolutional Neural Network

Phase drift extremely limits the secure key rate and secure transmission distance, which is non-negligible in local oscillation continuous variable quantum key distribution (LLO CV-QKD). In order to eliminate the impact caused by phase drift, we analyze the phase noise of the system and propose a phase compensation method based on convolutional neural network (CNN). Moreover, the compensation is performed on the signal according to the estimated value of phase drift before coherent detection. In numerical simulation, we compare the performance of phase compensation methods based on CNN and Kalman filter (KF), and the results show that CNN-based phase compensation has higher accuracy and stability.

KerGNNs: Interpretable Graph Neural Networks with Graph Kernels

Graph kernels are historically the most widely-used technique for graph classification tasks. However, these methods suffer from limited performance because of the hand-crafted combinatorial features of graphs. In recent years, graph neural networks (GNNs) have become the state-of-the-art method in downstream graph-related tasks due to their superior performance. Most GNNs are based on Message Passing Neural Network (MPNN) frameworks. However, recent studies show that MPNNs can not exceed the power of the Weisfeiler-Lehman (WL) algorithm in graph isomorphism test. To address the limitations of existing graph kernel and GNN methods, in this paper, we propose a novel GNN framework, termed Kernel Graph Neural Networks (KerGNNs), which integrates graph kernels into the message passing process of GNNs. Inspired by convolution filters in convolutional neural networks (CNNs), KerGNNs adopt trainable hidden graphs as graph filters which are combined with subgraphs to update node embeddings using graph kernels. In addition, we show that MPNNs can be viewed as special cases of KerGNNs. We apply KerGNNs to multiple graph-related tasks and use cross-validation to make fair comparisons with benchmarks. We show that our method achieves competitive performance compared with existing state-of-the-art methods, demonstrating the potential to increase the representation ability of GNNs. We also show that the trained graph filters in KerGNNs can reveal the local graph structures of the dataset, which significantly improves the model interpretability compared with conventional GNN models.

Deduction learning for precise noninvasive measurements of blood glucose with a dozen rounds of data for model training

Personalized modeling has long been anticipated to approach precise noninvasive blood glucose measurements, but challenged by limited data for training personal model and its unavoidable outlier predictions. To overcome these long-standing problems, we largely enhanced the training efficiency with the limited personal data by an innovative Deduction Learning (DL), instead of the conventional Induction Learning (IL). The domain theory of our deductive method, DL, made use of accumulated comparison of paired inputs leading to corrections to preceded measured blood glucose to construct our deep neural network architecture. DL method involves the use of paired adjacent rounds of finger pulsation Photoplethysmography signal recordings as the input to a convolutional-neural-network (CNN) based deep learning model. Our study reveals that CNN filters of DL model generated extra and non-uniform feature patterns than that of IL models, which suggests DL is superior to IL in terms of learning efficiency under limited training data. Among 30 diabetic patients as our recruited volunteers, DL model achieved 80% of test prediction in zone A of Clarke Error Grid (CEG) for model training with 12 rounds of data, which was 20% improvement over IL method. Furthermore, we developed an automatic screening algorithm to delete low confidence outlier predictions. With only a dozen rounds of training data, DL with automatic screening achieved a correlation coefficient ({R}_{P}) of 0.81, an accuracy score ({R}_{A}) of 93.5, a root mean squared error of 13.93 mg/dl, a mean absolute error of 12.07 mg/dl, and 100% predictions in zone A of CEG. The nonparametric Wilcoxon paired test on {R}_{A} for DL versus IL revealed near significant difference with p-value 0.06. These significant improvements indicate that a very simple and precise noninvasive measurement of blood glucose concentration is achievable.

Hybrid machine learning method for a connectivity-based epilepsy diagnosis with resting-state EEG.

This study investigates the performance of a convolutional neural network (CNN) algorithm on epilepsy diagnosis. Without pathology, diagnosis involves long and costly electroencephalographic (EEG) monitoring. Novel approaches may overcome this by comparing brain connectivity using graph metrics. This study, however, uses deep learning to learn connectivity patterns directly from easily acquired EEG data. A CNN algorithm was applied on directed Granger causality (GC) connectivity measures, derived from 50s of resting-state surface EEG recordings from 30 subjects with epilepsy and a 30 subject control group. The trained CNN filters reflected reduced delta band connectivity in frontal regions and increased left lateralized frontal-posterior gamma band connectivity. A diagnosis accuracy of 85% (F1 score 85%) was achieved by an ensemble of CNN models, each trained on differently prepared data from different electrode combinations. Appropriate preparation of connectivity data enables generic CNN algorithms to be used for detection of multiple discriminative epileptic features. Differential patterns revealed in this study may help to shed light on underlying altered cognitive abilities in epilepsy patients. The accuracy achieved in this study shows that, in combination with other methods, this approach could prove a valuable clinical decision support system for epilepsy diagnosis. 1: EEG measurements and subsequent connectivity calculation, 2: training of a neural network on resulting connectivity matrices, 3: extraction of most efficient CNN filters, which are neuromarker for epilepsy.

An Efficient Low-Complexity Convolutional Neural Network Filter

Convolutional neural network (CNN) filters have achieved significant performance in video artifacts reduction. However, the high complexity of existing methods makes them difficult to be applied in actual usage. In this article, an efficient low-complexity CNN filter is proposed. We utilized depth separable convolution merged with the batch normalization as the backbone of our proposed CNN filter and presented a frame-level residual mapping (RM) to use one network to filter both intra- and intersamples. It is known that there will be an oversmoothing problem for the interframes if we directly use the filter trained with intrasamples. In this article, the proposed RM can effectively solve the oversmoothing problem. Besides, RM is flexible and can be combined with other learning-based filters. The experimental results show that our proposed method achieves a significant bjøntegaard-delta(BD)-rate reduction than H.265/high efficiency video coding. The experiments show that the proposed network achieves about 1.2% BD-rate reduction and 79.1% decrease in FLOPs than VR-CNN. Our performance is better with less complexity than the previous work. The measurement on H.266/versatile video coding and ablation studies also ensure the effectiveness of the proposed method.

QA-Filter: A QP-Adaptive Convolutional Neural Network Filter for Video Coding.

Convolutional neural network (CNN)-based filters have achieved great success in video coding. However, in most previous works, individual models were needed for each quantization parameter (QP) band, which is impractical due to limited storage resources. To explore this, our work consists of two parts. First, we propose a frequency and spatial QP-adaptive mechanism (FSQAM), which can be directly applied to the (vanilla) convolution to help any CNN filter handle different quantization noise. From the frequency domain, a FQAM that introduces the quantization step (Qstep) into the convolution is proposed. When the quantization noise increases, the ability of the CNN filter to suppress noise improves. Moreover, SQAM is further designed to compensate for the FQAM from the spatial domain. Second, based on FSQAM, a QP-adaptive CNN filter called QA-Filter that can be used under a wide range of QP is proposed. By factorizing the mixed features to high-frequency and low-frequency parts with the pair of pooling and upsampling operations, the QA-Filter and FQAM can promote each other to obtain better performance. Compared to the H.266/VVC baseline, average 5.25% and 3.84% BD-rate reductions for luma are achieved by QA-Filter with default all-intra (AI) and random-access (RA) configurations, respectively. Additionally, an up to 9.16% BD-rate reduction is achieved on the luma of sequence BasketballDrill. Besides, FSQAM achieves measurably better BD-rate performance compared with the previous QP map method.

Analysis on performances of the optimization algorithms in CNN speech noise attenuator

In this paper, we studied the effect of the optimization algorithm of weight coefficients on the performance of the CNN(Convolutional Neural Network) noise attenuator. This system improves the performance of the noise attenuation by a deep learning algorithm using the neural network adaptive predictive filter instead of using the existing adaptive filter. Speech is estimated from a single input speech signal containing noise using 64-neuron, 16-filter CNN filters and an error back propagation algorithm. This is to use the quasi-periodic nature of the voiced sound section of the voice signal. In this study, to verify the performance of the noise attenuator for the optimization, a test program using the Keras library was written and training was performed. As a result of simulation, this system showed the smallest MSE value when using the Adam algorithm among the Adam, RMSprop, and Adagrad optimization algorithms, and the largest MSE value in the Adagrad algorithm. This is because the Adam algorithm requires a lot of computation but it has an exellent ability to estimate the optimal value by using the advantages of RMSprop and Momentum SGD.

Structural Compression of Convolutional Neural Networks with Applications in Interpretability.

Deep convolutional neural networks (CNNs) have been successful in many tasks in machine vision, however, millions of weights in the form of thousands of convolutional filters in CNNs make them difficult for human interpretation or understanding in science. In this article, we introduce a greedy structural compression scheme to obtain smaller and more interpretable CNNs, while achieving close to original accuracy. The compression is based on pruning filters with the least contribution to the classification accuracy or the lowest Classification Accuracy Reduction (CAR) importance index. We demonstrate the interpretability of CAR-compressed CNNs by showing that our algorithm prunes filters with visually redundant functionalities such as color filters. These compressed networks are easier to interpret because they retain the filter diversity of uncompressed networks with an order of magnitude fewer filters. Finally, a variant of CAR is introduced to quantify the importance of each image category to each CNN filter. Specifically, the most and the least important class labels are shown to be meaningful interpretations of each filter.

Facial expression recognition with grid-wise attention and visual transformer

Facial Expression Recognition (FER) has achieved remarkable progress as a result of using Convolutional Neural Networks (CNN). Relying on the spatial locality, convolutional filters in CNN, however, fail to learn long-range inductive biases between different facial regions in most neural layers. As such, the performance of a CNN-based model for FER is still limited. To address this problem, this paper introduces a novel FER framework with two attention mechanisms for CNN-based models, and these two attention mechanisms are used for the low-level feature learning the high-level semantic representation, respectively. In particular, in the low-level feature learning, a grid-wise attention mechanism is proposed to capture the dependencies of different regions from a facial expression image such that the parameter update of convolutional filters in low-level feature learning is regularized. In the high-level semantic representation, a visual transformer attention mechanism uses a sequence of visual semantic tokens (generated from pyramid features of high convolutional layer blocks) to learn the global representation. Extensive experiments have been conducted on three public facial expression datasets, CK+, FER+, and RAF-DB. The results show that our FER-VT has achieved state-of-the-art performance on these datasets, especially with a 100% accuracy on CK + datasets without any extra training data.

Investigating particle track topology for range telescopes in particle radiography using convolutional neural networks

Background Proton computed tomography (pCT) and radiography (pRad) are proposed modalities for improved treatment plan accuracy and in situ treatment validation in proton therapy. The pCT system of the Bergen pCT collaboration is able to handle very high particle intensities by means of track reconstruction. However, incorrectly reconstructed and secondary tracks degrade the image quality. We have investigated whether a convolutional neural network (CNN)-based filter is able to improve the image quality. Material and methods The CNN was trained by simulation and reconstruction of tens of millions of proton and helium tracks. The CNN filter was then compared to simple energy loss threshold methods using the Area Under the Receiver Operating Characteristics curve (AUROC), and by comparing the image quality and Water Equivalent Path Length (WEPL) error of proton and helium radiographs filtered with the same methods. Results The CNN method led to a considerable improvement of the AUROC, from 74.3% to 97.5% with protons and from 94.2% to 99.5% with helium. The CNN filtering reduced the WEPL error in the helium radiograph from 1.03 mm to 0.93 mm while no improvement was seen in the CNN filtered pRads. Conclusion The CNN improved the filtering of proton and helium tracks. Only in the helium radiograph did this lead to improved image quality.

Pruning CNN filters via quantifying the importance of deep visual representations

The achievement of convolutional neural networks (CNNs) in a variety of applications is accompanied by a dramatic increase in computational costs and memory requirements. In this paper, we propose a novel framework to measure the importance of individual hidden units by computing a measure of relevance to identify the most critical filters and prune them to compress and accelerate CNNs. Unlike existing methods, we introduce the use of the activation of feature maps to detect valuable information and the essential semantic parts to evaluate the importance of feature maps, inspired by novel neural network interpretability. A majority voting technique based on the degree of alignment between a semantic concept and individual hidden unit representations is proposed to quantitatively evaluate the importance of feature maps. We also propose a simple yet effective method to estimate new convolution kernels based on the remaining, crucial channels to accomplish effective CNN compression. Experimental results show the effectiveness of our filter selection criteria, which outperforms the state-of-the-art baselines. Furthermore, we evaluate our pruning method on CIFAR-10, CUB-200, and ImageNet (ILSVRC 2012) datasets. The experimental results show that the proposed method efficiently achieves a 50% FLOPs reduction on CIFAR-10, with only 0.86% accuracy drop on the VGG-16 model. Meanwhile, ResNet pruned on CIFAR-10 achieves a 30% reduction in FLOPs with only 0.12% and 0.02% drops in accuracy on ResNet-20 and ResNet-32 respectively. For ResNet-50 on ImageNet, our pruned model achieves a 50% reduction in FLOPs with only a top-5 accuracy drop of 0.27%, which significantly outperforms state-of-the-art methods.