Image Classification Research Articles

Active self-semi-supervised learning for few labeled samples

Training deep models with limited annotations poses a significant challenge when applied to diverse practical domains. Employing semi-supervised learning alongside the self-supervised model offers the potential to enhance label efficiency. However, this approach faces a bottleneck in reducing the need for labels. We observed that the semi-supervised model disrupts valuable information from self-supervised learning when only limited labels are available. To address this issue, this paper proposes a simple yet effective framework, active self-semi-supervised learning (AS3L). AS3L bootstraps semi-supervised models with prior pseudo-labels (PPL). These PPLs are obtained by label propagation over self-supervised features. Based on the observations the accuracy of PPL is not only affected by the quality of features but also by the selection of the labeled samples. We develop active learning and label propagation strategies to obtain accurate PPL. Consequently, our framework can significantly improve the performance of models in the case of limited annotations while demonstrating fast convergence. On the image classification tasks across four datasets, our method outperforms the baseline by an average of 5.4%. Additionally, it achieves the same accuracy as the baseline method in about 1/3 of the training time.

Evaluation of super resolution technology for digestive endoscopic images

ObjectThis study aims to evaluate the value of super resolution (SR) technology in augmenting the quality of digestive endoscopic images. MethodsIn the retrospective study, we employed two advanced SR models, i.e., SwimIR and ESRGAN. Two discrete datasets were utilized, with training conducted using the dataset of the First Affiliated Hospital of Soochow University (12,212 high-resolution images) and evaluation conducted using the HyperKvasir dataset (2,566 low-resolution images). Furthermore, an assessment of the impact of enhanced low-resolution images was conducted using a 5-point Likert scale from the perspectives of endoscopists. Finally, two endoscopic image classification tasks were employed to evaluate the effect of SR technology on computer vision (CV). ResultsSwinIR demonstrated superior performance, which achieved a PSNR of 32.60, an SSIM of 0.90, and a VIF of 0.47 in test set. 90 % of endoscopists supported that SR preprocessing moderately ameliorated the readability of endoscopic images. For CV, enhanced images bolstered the performance of convolutional neural networks, whether in the classification task of Barrett's esophagus (improved F1-score: 0.04) or Mayo Endoscopy Score (improved F1-score: 0.04). ConclusionsSR technology demonstrates the capacity to produce high-resolution endoscopic images. The approach enhanced clinical readability and CV models’ performance of low-resolution endoscopic images.

Alignable kernel network

To enhance the adaptability and performance of Convolutional Neural Networks (CNN), we present an adaptable mechanism called Alignable Kernel (AliK) unit, which dynamically adjusts the receptive field (RF) dimensions of a model in response to varying stimuli. The branches of AliK unit are integrated through a novel align transformation softmax attention, incorporating prior knowledge through rank ordering constraints. The attention weightings across the branches establish the effective RF scales, leveraged by neurons in the fusion layer. This mechanism is inspired by neuroscientific observations indicating that the RF dimensions of neurons in the visual cortex vary with the stimulus, a feature often overlooked in CNN architectures. By aggregating successive AliK ensembles, we develop a deep network architecture named the Alignable Kernel Network (AliKNet). AliKNet with interdisciplinary design improves the network’s performance and interpretability by taking direct inspiration from the structure and function of human neural systems, especially the visual cortex. Empirical evaluations in the domains of image classification and semantic segmentation have demonstrated that AliKNet excels over numerous state-of-the-art architectures, achieving this without increasing model complexity. Furthermore, we demonstrate that AliKNet can identify target objects across various scales, confirming their ability to dynamically adapt their RF sizes in response to the input data.

A Distance Transformation Deep Forest Framework With Hybrid-Feature Fusion for CXR Image Classification.

Detecting pneumonia, especially coronavirus disease 2019 (COVID-19), from chest X-ray (CXR) images is one of the most effective ways for disease diagnosis and patient triage. The application of deep neural networks (DNNs) for CXR image classification is limited due to the small sample size of the well-curated data. To tackle this problem, this article proposes a distance transformation-based deep forest framework with hybrid-feature fusion (DTDF-HFF) for accurate CXR image classification. In our proposed method, hybrid features of CXR images are extracted in two ways: hand-crafted feature extraction and multigrained scanning. Different types of features are fed into different classifiers in the same layer of the deep forest (DF), and the prediction vector obtained at each layer is transformed to form distance vector based on a self-adaptive scheme. The distance vectors obtained by different classifiers are fused and concatenated with the original features, then input into the corresponding classifier at the next layer. The cascade grows until DTDF-HFF can no longer gain benefits from the new layer. We compare the proposed method with other methods on the public CXR datasets, and the experimental results show that the proposed method can achieve state-of-the art (SOTA) performance. The code will be made publicly available at https://github.com/hongqq/DTDF-HFF.

A Survey of Automated Data Augmentation for Image Classification: Learning to Compose, Mix, and Generate.

Data augmentation is an effective way to improve the generalization of deep learning models. However, the underlying augmentation methods mainly rely on handcrafted operations, such as flipping and cropping for image data. These augmentation methods are often designed based on human expertise or repeated trials. Meanwhile, automated data augmentation (AutoDA) is a promising research direction that frames the data augmentation process as a learning task and finds the most effective way to augment the data. In this survey, we categorize recent AutoDA methods into the composition-, mixing-, and generation-based approaches and analyze each category in detail. Based on the analysis, we discuss the challenges and future prospects as well as provide guidelines for applying AutoDA methods by considering the dataset, computation effort, and availability of domain-specific transformations. It is hoped that this article can provide a useful list of AutoDA methods and guidelines for data partitioners when deploying AutoDA in practice. The survey can also serve as a reference for further study by researchers in this emerging research area.

Adaptive prototype few-shot image classification method based on feature pyramid

Few-shot learning aims to enable machines to recognize unseen novel classes using limited samples akin to human capabilities. Metric learning is a crucial approach to addressing this challenge, with its performance primarily dependent on the effectiveness of feature extraction and prototype computation. This article introduces an Adaptive Prototype few-shot image classification method based on Feature Pyramid (APFP). APFP employs a novel feature extraction method called FResNet, which builds upon the ResNet architecture and leverages a feature pyramid structure to retain finer details. In the 5-shot scenario, traditional methods for computing average prototypes exhibit limitations due to the typically diverse and uneven distribution of samples, where simple means may inadequately reflect such diversity. To address this issue, APFP proposes an Adaptive Prototype method (AP) that dynamically computes class prototypes of the support set based on the similarity between support set samples and query samples. Experimental results demonstrate that APFP achieves 67.98% and 85.32% accuracy in the 5-way 1-shot and 5-way 5-shot scenarios on the MiniImageNet dataset, respectively, and 84.02% and 94.44% accuracy on the CUB dataset. These results indicate that the proposed APFP method addresses the few-shot learning problem.

Deep convolutional neural networks for ship detection using refined DOTA and TGRS-HRRSD high-resolution image datasets

Automated identification of ships in satellite imagery is of utmost significance in many military and civilian applications, including monitoring maritime traffic and maintaining maritime security. However, ship detection from remotely sensed imagery is a challenging task because of their varying sizes, orientations, types, and various interferences. In recent years, deep learning algorithms have become a powerful tool in image processing applications, including image classification and object detection due to their strong feature representation capabilities. In this study, a refined ship dataset, containing a total of 13,735 instances and representing different ship types, was generated using state-of-the-art datasets for the ship detection task. In a comparative framework, the performance of five deep learning algorithms (Faster R-CNN, YOLOv8, CenterNet, EfficientDet, and SSD) in ship detection was investigated using the created dataset. The performance of the deep learning models was compared using MS COCO accuracy metrics. In addition, the generalization capabilities of the models were tested on an independent image acquired by Göktürk-1 having a mucilage effect that undermines the spectral discrimination. Results revealed that the YOLOv8 model had the greatest ship detection performance (AP@0.50 = 96.33 %), followed by CenterNet (AP@0.50 = 88.80 %), Faster R-CNN (AP@0.50 = 85.30 %), EfficientDet (AP@0.50 = 78.91 %) and SSD (AP@0.5 = 75.98 %), and obtained the highest accuracies in terms of other COCO metrics. In addition, generalization capabilities on the test image confirmed the superiority of the YOLOv8 model with AP@0.50 score of 63.93 %. Overall, the results of this study validated the effectiveness of the YOLOv8 model in identifying ship targets in remotely sensed images.

BI‐LAVA: Biocuration With Hierarchical Image Labelling Through Active Learning and Visual Analytics

AbstractIn the biomedical domain, taxonomies organize the acquisition modalities of scientific images in hierarchical structures. Such taxonomies leverage large sets of correct image labels and provide essential information about the importance of a scientific publication, which could then be used in biocuration tasks. However, the hierarchical nature of the labels, the overhead of processing images, the absence or incompleteness of labelled data and the expertise required to label this type of data impede the creation of useful datasets for biocuration. From a multi‐year collaboration with biocurators and text‐mining researchers, we derive an iterative visual analytics and active learning (AL) strategy to address these challenges. We implement this strategy in a system called BI‐LAVA—Biocuration with Hierarchical Image Labelling through Active Learning and Visual Analytics. BI‐LAVA leverages a small set of image labels, a hierarchical set of image classifiers and AL to help model builders deal with incomplete ground‐truth labels, target a hierarchical taxonomy of image modalities and classify a large pool of unlabelled images. BI‐LAVA's front end uses custom encodings to represent data distributions, taxonomies, image projections and neighbourhoods of image thumbnails, which help model builders explore an unfamiliar image dataset and taxonomy and correct and generate labels. An evaluation with machine learning practitioners shows that our mixed human–machine approach successfully supports domain experts in understanding the characteristics of classes within the taxonomy, as well as validating and improving data quality in labelled and unlabelled collections.

LoVi App: Android Application-based Image Classification for Low Vision

In Indonesia, many people with visual impairments are drawing public attention to their rights as fellow humans. One of the limitations that individuals with low vision face is their ability to recognize objects and navigate their surroundings due to difficulties in visual perception. In this modern era, deep learning technologies, especially in image classification, can help people with low vision overcome these challenges. In this paper, we discuss a deep learning system that optimizes image classification on users' smartphones to enhance visual support for individuals with low vision. We present an Android-based app, LoVi, designed to assist users with low vision. Powered by core systems within Sherpa models (TrotoarNet, IndoorNet, and CurrencyNet), LoVi has three modes: outdoor, indoor, and currency. The LoVi application provides over 80% accuracy for navigation on sidewalks, indoor object recognition, and currency identification. TrotoarNet aids in sidewalk navigation, IndoorNet assists with indoor object identification, and CurrencyNet recognizes Rupiah banknotes. Additionally, low-vision users can receive voice feedback for further accessibility.

Transformer based on multi-scale local feature for colon cancer histopathological image classification

Colon cancer histopathological image recognition is an important method in the early screening of colon cancer. At present, most whole slide images (WSIs) classification models only consider the features of a single patch, and do not consider the features of the WSIs region. To expand the receptive field of the patches and obtain its surrounding pathological information, this paper proposes a transformer based on multi-scale local feature for colon cancer histopathological image classification model. The model samples patches based on farthest point sampling (FPS). A multi-scale grouping scheme is also proposed to capture WSI local regions. Then the classification labels, feature information and local area information of sampled patches are embedded into the transformer encoder. In addition, the focal loss function is adopted to solve the problem of unbalanced categories of colon cancer histopathological images. Experimental results demonstrate that our model surpasses competing methods in performance on both hospital colon cancer datasets and The Cancer Genome Atlas(TCGA) colon cancer datasets.

Optimizing Deep Learning Parameters for Enhanced Image Classification Accuracy: A Theoretical and Empirical Analysis

This study investigates the,impact of key deep learning,parameters—algorithm,type, dataset size, learning,rate, batch size, and number,of epochs—on image,classification accuracy, a,critical concern,in industries such as,healthcare, security, and,technology. Despite,advancements in deep learning, optimizing,these parameters for maximum,accuracy remains,challenging. To address,this issue, a,survey was conducted,among 381 professionals and,academics experienced,in deep learning, and,the data was analyzed,using Partial Least Squares,Structural Equation,Modeling (PLS-SEM). The,results confirm that all,five parameters significantly,influence image classification,accuracy, with optimized,settings leading to substantial,improvements. This,study contributes to,the theoretical understanding,of the Bias-Variance,Tradeoff Theory in the,context of deep learning,and offers practical guidelines,for enhancing model,performance. However, limitations,such as reliance on,self-reported data suggest,the need for further,experimental research. The,findings provide a comprehensive,framework for optimizing,deep learning models, offering,both academic insights and,practical solutions for,improving accuracy in,critical applications.

Whisperings From The Ordinary: A Conversational Artificial Intelligence (AI) Approach To Bridging The Human-Object Divide

In a world increasingly dominated by connected devices, one realm remains silent: most inanimate objects that surround us. This stark contrast highlights a critical gap in human-environment interaction, one that the present work aims to bridge through the power of intelligent conversation and shared understanding. Traditional IoT methods, while efficient, raise crucial concerns regarding sustainability. Embedding electronics into every object risk exceeding costs and generating substantial e-waste. This necessitates exploration of alternative paradigms that unlock the communicative potential of everyday items without environmental burden. Enter the confluence of object detection and conversational AI. Object detection, encompassing techniques like image classification, empowers machines to perceive their surroundings and make informed decisions based on identified objects. Conversational AI refers to the ability of computers to engage in a natural, human-like dialogue. By fitting in object detection, conversational AI can transform lifeless objects into active and intelligent participants in our lives, fostering a richer relationship with the spaces we inhabit. This paper proposes an AI-powered mobile application that unlocks the communicative potential of everyday items, transforming them from passive elements into active participants in our lives. We are exploring to enhance domestic routines by streamlining daily tasks with object-initiated reminders, proactive maintenance suggestions, and personalized advice. This research delves into the technical aspects, potential applications, and ethical implications of granting voice to the voiceless.

Classification of Malaria Using Convolutional Neural Network Method on Microscopic Image of Blood Smear

Malaria, a critical global health issue, can lead to severe complications and mortality if not treated promptly. The conventional diagnostic method, involving a microscopic examination of blood smears, is time-consuming and requires extensive expertise. To address these challenges, computer-assisted diagnostic methods have been explored. Among these, Convolutional Neural Networks (CNN), a deep learning technique, has shown considerable promise for image classification tasks, including the analysis of microscopic blood smear images. In this study, we employ the NIH Malaria dataset, which consists of 27,558 images, to train a CNN model. The dataset is divided into parasitized (malaria-infected) and uninfected. The CNN architecture designed for this study includes three convolutional layers and two fully connected layers. We compare the performance of this model with that of a pre-trained VGG-16 model to determine the most effective approach for malaria diagnosis. The proposed CNN model demonstrates high accuracy, achieving a value of 96.81%. Furthermore, it records a recall of 0.97, a precision of 0.97, and an F1-score of 0.97. These metrics indicate a robust performance, outperforming previous studies and highlighting the model's potential for accurate malaria diagnosis. This study underscores the potential of CNN in medical image classification and supports its implementation in clinical settings to enhance diagnostic accuracy and efficiency. The findings suggest that with further refinement and validation, such models could significantly improve the speed and reliability of malaria diagnostics, ultimately aiding in better disease management and patient outcomes.

Blending Ensemble applied to Open-Set Recognition for Time Series Classification

Time Series (TS), including stock prices, temperatures, and health markers are monitored all the time and everywhere. Time-dependent events, and TS are frequently studied in the scope of forecasting, where past values of the TS are used to preview future ones. On the other hand, Time Series Classification (TSC) aims at creating models that label TS instances. Open Set Recognition (OSR) techniques, designed to classify known samples and detect unknowns simultaneously, have been less applied in TSC compared to other fields like image classification. Existing methods have limitations, such as not using known-unknown in the training stage, limited transfer learning across Neural Network models, and experiments with benchmark datasets with narrow coverage. This study introduces a novel OSR approach for TSC by addressing the mentioned issues and using a blending ensemble of Neural Networks with an OpenMax layer. The results vouch for the model’s performance and potential superiority as an alternative to existing methods in open-set recognition tasks.

Classification of BPG-Based Lossy Compressed Noisy Images

Acquired remote sensing images can be noisy. This fact has to be taken into account in their lossy compression and classification. In particular, a specific noise filtering effect is usually observed due to lossy compression and this can be positive for classification. Classification can be also influenced by methodology of classifier learning. In this paper, we consider peculiarities of lossy compression of three-channel noisy images by better portable graphics (BPG) encoder and their further classification. It is demonstrated that improvement of data classification accuracy is not observed if a given image is compressed in the neighborhood of optimal operation point (OOP) and the classifier training is performed for the noisy image. Performance of neural network based classifier is studied. As demonstrated, its training for compressed remote sensing data is able to provide certain benefits compared to training for noisy (uncompressed) data. Examples for Sentinel data used in simulations are offered.

Computationally optimized brain tumor classification using attention based GoogLeNet-style CNN

Early detection of diseases is a crucial step towards patient healthcare and recovery, especially, for low survival rate diseases like brain cancer. In recent years, deep learning models such as convolutional neural networks (CNNs) have been widely utilized for medical image classification owing to their proficiency in extracting local features. However, accurate classification of complex medical images necessitates capturing local and global features. This paper focuses on the classification of the three most prominent kinds of brain tumors, namely, glioma, meningioma and pituitary tumor, with the proposed CNN based on the structure of GoogLeNet. The proposed model uses an attention-based inception module (ABIM) embedded with attention mechanism of spatial attention and efficient channel attention. Attention helps CNN focus on features with more distinguishing power and ignores irrelevant features while the inception approach aids in extracting multiscale local features. ABIM incorporates attention with inception to generate better feature representation. The model training and testing are done on a publicly accessible Figshare dataset with 3064 brain MRI samples. The dataset comprises 1426, 708, and 930 MRI samples of glioma, meningioma, and pituitary tumors, respectively. In a five-fold cross-validation experiment, the proposed model achieved an average accuracy of 97.97% for glioma, 94.76% for meningioma, and 99.22% for pituitary tumors. The proposed CNN model surpasses existing models with an overall accuracy of 97.62%, while utilizing fewer parameters than its base model, GoogLeNet. The model is also evaluated on datasets from the Kaggle and Harvard medical repositories, and showed promising results in distinguishing between different tumor classes, and normal brain MRIs. These results demonstrate that our model not only outperforms other existing models in terms of accuracy but also has comparatively lower computational complexity, providing a robust tool to support healthcare professionals.

SSResNeXt: A Novel Deep Learning Architecture for Multi-class Breast Cancer Pathological Image Classification

Multi-class classification of breast cancer pathological images remains challenging due to complex image features and limited datasets. This study proposes SSResNeXt, a novel deep learning architecture incorporating a new Small-SE-ResNeXt Block with asymmetric convolutions and channel attention mechanisms. Evaluated on the BreaKHis dataset, SSResNeXt achieves state-of-the-art performance with accuracies of 95.2%, 94.0%, 92.7%, and 93.5% for 40X, 100X, 200X, and 400X magnification scales, respectively. Comparative experiments demonstrate SSResNeXt's superior performance over existing models, including ResNet and Swin Transformer variants. The proposed architecture offers improved feature extraction capabilities for breast cancer pathological images without significantly increasing model complexity, providing a promising tool for computer-aided diagnosis systems.

Deep Learning-Based Leafy Vegetable Image Classification for Smart Farm Automation

Deep Learning-Based Leafy Vegetable Image Classification for Smart Farm Automation

Using machine learning models to detect artificially generated images

The use of machine learning models to classify real and artificially generated images is becoming an increasingly relevant area of research in the field of artificial intelligence. This paper is dedicated to the analysis and development of various machine learning models used for this task. Our main goal is to investigate the effectiveness of different neural network architectures in classifying images that have been created both in by human and with the help of artificial intelligence. As part of the study, we used nine different neural network architectures designed to analyse their effectiveness in classifying images. We studied both real images and those generated by artificial intelligence. The latter are becoming more and more common in the modern world, which means that there is a need for their identification and classification. The main results of our study include a detailed analysis of the characteristics of the generated images, as well as a comparison of the models using various metrics. We used metrics such as accuracy and F1-score, which allowed us to objectively evaluate the performance of each model. In addition, we identified the most effective models for the image classification task. One of the key findings of our study is that models that use data regularisation and augmentation performed better. Data regularisation and augmentation help to ensure more stable classification accuracy and reduce the tendency to overfitting, which is an important issue in machine learning. These results can be useful for developing strategies to counteract disinformation, which has become an issue in the modern information society. By using machine learning models, artificially generated images can be detected and separated from real images, which can help prevent the spread of fake news and disinformation. In addition, the results of our research can contribute to the further improvement of image generation algorithms, which is an important area of artificial intelligence development. They can serve as a basis for creating more accurate and efficient models that can generate images that are as close to real life as possible.

Editorial for Vol.32, No.2

The June 2024 (Vol. 32, No. 2) issue of CIT. Journal of Computing and Information Technology brings four papers from the areas of natural language processing, image processing, network security, and biomedical image classification.