Class noise removal and correction for image classification using ensemble margin

Noise correction to improve data and model quality for crowdsourcing

The accuracy of a classifier, whether it is an ensemble or not, is directly influenced by the training data used in learning. In remote sensing, training data mislabeling is inevitable and faces a major challenge. This paper proposes a versatile data cleaning which handles the mislabeling problem by exploiting the ensemble concepts for identifying, then eliminating or correcting the mislabeled training data. A powerful ensemble method, random forest, is at the core of our filter design and helps to distinguish mislabeled data from uncorrupted data more accurately. The major contribution of this work lies on the explicit use of the hypothesis margin as a decision means to identify and eliminate or correct mislabeled training data in an ensemble learning framework. Another key development that makes our algorithm superior to existing approaches is a design that avoids rare class instances to be mistaken for class noise. This fundamental aspect makes our data cleaning system particularly suitable for remote sensing classification tasks which usually suffer from both mislabeling and imbalance problems. The effectiveness of our algorithm is demonstrated in performing mapping of land covers. The generalization performance of two major supervised noise-sensitive classifiers, boosting and K-nearest neighbors, is strengthened by effective class noise reduction. A comparative analysis is conducted with respect to random forest, deep convolutional neural networks, as well as two well-established ensemble-based class noise filters, the majority vote and the consensus vote filters. This analysis demonstrates that our approach is more accurate than deep convolutional neural networks (one-dimensional CNN, AlexNet, EfficientNet, ResNet50 and ShuffletNet) and the reference ensemble methods.

Introduction: Functional MRI methods have been used to study sensorimotor processing in the brain and the Spinal cord. However, these techniques confront unwanted contributions to the measured signal from physiological fluctuations. For the spinal cord imaging, most of the challenges are consequences of cardiac and respiratory movement artifacts that are considered as significant sources of noise. Spinal cord in Thoraco-Lumbar is close to lungs and diaphragms that influence cerebrospinal fluid-filled spaces and cause changes in thesusceptibility due to the change in the amount of air in the lung. In this study, we investigatedthe effect of each source of physiological noise and contribution of them to the outcome ofthe analysis of the BOLD signal in human spinal cord during a sensory stimulation of the foot. Materials and Methods: Fifteen young healthy male volunteers participated in the study. Sixty pressure pain stimuli delivered on L5 dermatome between the two malleoli. All of functional data were collected using a 3T Siemens Prisma scanner with 31 axial slices T2*-weighted ZOOMit (TR = 3000 ms; TE = 30 ms; FA = 90°; FoV = 160 × 160 mm; matrix size = 64 × 64). Respiratory and cardiac signals were recorded during the imaging session using the data acquired from theimplemented physiological monitoring unit. The Spinal Cord Toolbox and FSL were used forimage processing and analysis. Generated respiration and cardiac regressors were includedin the GLM for the quantification of the effect of each of them on the task-analyses results. The sum of activated voxels of the clusters in the spinal cord and in all the image was calculated. Results: Comparison of sum of activated voxels. A one-way within subject’s ANOVA was applied toevaluate the effect of physiological functions on spinal cord fMRI in three noise correction inthe GLM. There was a statistically significant effect of physiological noise correctiononnumber of activated voxels in Image, F (3, 42) = 3.00689, p = .040817, η2 =0.17, and thisvalues in spinal cord was F(3, 42) = 21.314, p =.00001, , η2 =0.6. Bonferroni post hoc testsillustrate that cardiac noise correction had an effective role on increase activated voxels (mean = 23.46±9.46) compared to other noise correction methods. Conclusion: The effect of the Heart-beat and Respiration movements has a significant role in thephysiological noise and concurrency between movements with task and correction of thismovements can remove the real effect of stimulation. Cardiac effect and changes in thecorrected results are higher than other physiological noise sources. In spite of previous work,displacement of the spinal cord and effect of this noise in the fMRI results are significant andcannot be ignored.

Objective. Learning to classify cardiac abnormalities requires large and high-quality labeled datasets, which is a challenge in medical applications. Small datasets from various sources are often aggregated to meet this requirement, resulting in a final dataset prone to label noise due to inter- and intra-observer variability and different expertise. It is well known that label noise can affect the performance and generalizability of the trained models. In this work, we explore the impact of label noise and self-learning label correction on the classification of cardiac abnormalities on large heterogeneous datasets of electrocardiogram (ECG) signals. Approach. A state-of-the-art self-learning multi-class label correction method for image classification is adapted to learn a multi-label classifier for electrocardiogram signals. We evaluated our performance using 5-fold cross-validation on the publicly available PhysioNet/Computing in Cardiology (CinC) 2021 Challenge data, with full and reduced sets of leads. Due to the unknown label noise in the testing set, we tested our approach on the MNIST dataset. We investigated the performance under different levels of structured label noise for both datasets. Main results. Under high levels of noise, the cross-validation results of self-learning label correction show an improvement of approximately 3% in the challenge score for the PhysioNet/CinC 2021 Challenge dataset and an improvement in accuracy of 5% and reduction of the expected calibration error of 0.03 for the MNIST dataset. We demonstrate that self-learning label correction can be used to effectively deal with the presence of unknown label noise, also when using a reduced number of ECG leads.

Image classification is getting more attention in the area of computer vision. During the past few years, a lot of research has been done on image classification using classical machine learning and deep learning techniques. Presently, deep learning-based techniques have given stupendous results. The performance of a classification system depends on the quality of features extracted from an image. The better is the quality of extracted features, the more the accuracy will be. Although, numerous deep learning-based methods have shown enormous performance in image classification, still due to various challenges deep learning methods are not able to extract all the important information from the image. This results in a reduction in overall classification accuracy. The goal of the present research is to improve the image classification performance by combining the deep features extracted using popular deep convolutional neural network, VGG19, and various handcrafted feature extraction methods, i.e., SIFT, SURF, ORB, and Shi-Tomasi corner detector algorithm. Further, the extracted features from these methods are classified using various machine learning classification methods, i.e., Gaussian Naïve Bayes, Decision Tree, Random Forest, and eXtreme Gradient Boosting (XGBClassifier) classifier. The experiment is carried out on a benchmark dataset Caltech-101. The experimental results indicate that Random Forest using the combined features give 93.73% accuracy and outperforms other classifiers and methods proposed by other authors. The paper concludes that a single feature extractor whether shallow or deep is not enough to achieve satisfactory results. So, a combined approach using deep learning features and traditional handcrafted features is better for image classification.

This research paper presents an advanced deep learning framework for tomato ripeness and maturity detection, employing Convolutional Neural Networks (CNNs) as the primary tool for automated classification. In particular, the study emphasizes the utilization of the ResNet-152 architecture, a sophisticated deep neural network known for its exceptional performance in image classification tasks. ResNet-152 addresses the challenges of fine-grained categorization by enabling the extraction of intricate visual features such as color nuances, textures, and shapes inherent to tomatoes. This eliminates the necessity for manual feature engineering, enhancing the model's ability to discern between "RIPE," "UNRIPE," and "ROTTEN" tomato classes with unprecedented accuracy. The ResNet-152 architecture's effectiveness lies in its unique design, featuring residual blocks that facilitate the training of very deep networks. This architecture mitigates the vanishing gradient problem and enables the model to learn complex hierarchical features, contributing to its state-of-the-art performance in image classification. In the context of tomato ripeness detection, ResNet-152 acts as a powerful tool for capturing and understanding the intricate visual cues that define different ripeness states, laying the foundation for an efficient and accurate automated sorting system in the agricultural and food processing industries.

In the realm of computer vision, image classification is a critical issue with many applications, including multimedia content analysis, security and surveillance, and medical imaging. The accuracy of image classification algorithms has considerably increased with the development of deep learning. The discipline of image classification has dramatically benefited from the usage of pre-trained Convolutional Neural Network (CNN) models. In this study, we perform image classification on the Wang dataset composed of 1000 images separated into ten categories, using six different pre-trained CNN models. The research made use of pre-trained versions of VGG16, Densenet, Mobilenet, Inception V3, Resnet50, and Xception models. We assessed the model performances in terms of training and testing accuracy. Testing accuracies for ten unique batches of data were calculated and averaged out. Densenet outperformed state-of-the-art models like Xception by a small margin, mainly due to low data availability. The findings of this study show how pre-trained models have advanced the field of image classification and shed light on how well these models perform in diverse image classification tasks. This study can serve as a valuable reference for researchers and practitioners in the field of computer vision and deep learning and help inform their choices when selecting pre-trained models for image classification, depending on their needs, while also considering data availability and computational constraints.

Image classification toward breast cancer using deeply-learned quality features

Bag of visual words model deriving from text categorization has recently appeared promising for object and image classification, this method always need to deal with large database. This paper proposed an efficient clustering algorithm to obtain universal codebook and adapted codebook, our combination of k-means and agglomerative clustering gives significant improvement in time efficiency while maintaining the same performance of image classification. We also use the adapted codebook to improve image classification performance, an image is presented by a set of histograms - one per class, each histogram describes whether the image is best modeled by the universal codebook or the corresponding adapted class codebook. The experiment result on Caltech-256 shows the combined universal codebook and adapted class codebook representation outperforms those approaches which use the universal codebook only.

With the development of deep learning, CNNs have become mainstream in image classification. However, CNN performance is affected by factors like loss function choice and optimization algorithm. This paper aims to improve CNN performance in image classification by optimizing SGD. Firstly, we analyze different loss functions and propose a suitable optimization strategy. Second, we explore SGD and its variants' roles in CNN training, focusing on accelerating convergence and improving accuracy. Experiments show the proposed scheme effectively enhances CNN performance, providing a reliable solution for image classification. This study supports CNN applications in image recognition and lays a foundation for exploring more efficient optimization methods.

Multitasking environments, such as monitoring of multiple underwater drones during live rapid environmental assessment (REA), while performing image classification, require effective cognitive resource management. In this second of two papers, we examine how trust, task-switching demands, workload, auditory environment, affect image classification task performance. We also benchmark a novel machine learning classification algorithm (Bijjahalli et al., 2023) against human performance. Results of our Bayesian multilevel modelling (n=95) found trust in reliable alerts supported task focus but had limited impact on response times. Task-switching interruptions led to a decline in immediate performance, which improved over time, and that image classification accuracy was unaffected by perceived workload, although moderate increases in effort and frustration improved accuracy. We also found that silent operating environments significantly improved image classification accuracy, while female voice alerts were distracting. Machine learning predictions aligned with faster response times and accuracy, supporting their utility in a decision support system.

Image classification is a very common research area, on which researchers work with various classification techniques. The aim of this study is to apply different filters on four different datasets and evaluate their performances in image classification. The study was performed in WEKA environment with Random Forest algorithm and image filters are applied to the datasets one by one and as a combination. Filter combinations got better performance than applying single filter on data. Filter combinations got the worst result on artworks with a percentage of 83.42%. However they were very successful on classifying the images in natural images dataset with a performance of 99.76%.

Image classification is an important task that has a wide range of applications. Convolutional neural networks (CNNs) are a common approach that can achieve promising performance in image classification. However, using CNNs to address a problem requires in-depth knowledge about CNN architectures and how it relates to the problem domain. Genetic programming (GP) as an evolutionary computation method can used to reduce the amount of knowledge required to design a CNN for a given problem domain by automatically searching for the optimal architecture. This paper proposes a new algorithm named, GPCNN, which encodes graph-based CNN architectures as trees and uses genetic operators, i.e. mutation, crossover and selection, to find better architectures. A more flexible crossover, partial subtree crossover, is also proposed to improve the search performance. As an preliminary work, GPCNN did not manage to achieve better performance than the state-of-the-art methods due to the limit on computational resource, but it is able to achieve better results than the baseline methods. More importantly, the proposed tree-based graph representation of CNN allows CNN architecture of various shapes, which has a great potential for future work in evolutionary automatic neural architecture search.

With the development of the Internet of Things, the application of computer vision on mobile phones is becoming more and more extensive and people have higher and higher requirements for the timeliness of the recognition results returned and the processing capabilities of the mobile phone for image recognition. However, the processing capability and storage capability of the user terminal equipment cannot meet the needs of identifying and storing a large number of pictures, and the data transmission process will cause high energy consumption of the terminal equipment. At the same time, multisource deep transfer learning has outstanding performance in computer vision and image classification. However, due to the huge amount of calculation of the deep network model, it is impossible to use the existing excellent network model to realize image recognition and classification on the mobile terminal. In order to solve the abovementioned problems, we propose a multisource mobile transfer learning algorithm based on dynamic model compression, this algorithm considers the realization of multisource transfer learning computing in the case of multiple mobile device computing source domains, and the method also guarantees data privacy and security for each device (origin domain). Meanwhile, extensive experiments show that our method can achieve remarkable results in popular image classification datasets.

Skin cancer image classification is critical to improve healthcare outcomes. Current practice often involves time-consuming procedures that may delay diagnosis until the disease has progressed to an advanced stage, reducing the chances of successful treatment. This challenge is further exacerbated by the worldwide shortage of skilled dermatologists. In this study, we investigate the effect of dataset size on the image classification performance of eight networks (AlexNet, ResNet18, ResNet34, ResNet50, ResNet101, ResNet152, ViT, and MLP-Mixer). We trained these classifiers using different ratios (e.g. 1% to 100%) of samples from the HAM10000 dataset. Our experiment reveals the complex interplay among dataset size, model complexity, and skin cancer classification performance, validating the rules of the neural scaling laws on skin cancer image classification. This work highlights the impact of dataset scale and model complexity on improving the skin cancer image classification performance to potentially reduce the burden on healthcare professionals.