Automatic Image Classification Research Articles

TRANSFER LEARNING IN THE CLASSIFICATION OF SATELLITE IMAGES SHOWING AMAZON RAINFOREST

Abstract. In recent years, we have been dealing with the dynamic technological progress of the space sector, which allows for the observation of the Earth with better temporal, spatial and spectral resolution. The increasing availability of satellite data has contributed to the development of data processing algorithms. Thanks to the use of digital image processing methods and deep neural networks, it is possible to perform automatic image classification, segmentation or detection and recognition of objects on the images. This article presents the methodology that allows to accelerate the classification process of satellite images representing the Amazon rainforest based on the Transfer Learning method. Additionally, the influence of the choice of optimization, i.e. the network weight estimation strategy, on the classification of objects was checked. In order to verify the method, an additional raster image classifier was created on the basis of Lidar data. Research shows that the transfer learning method allows the preparation of an image classifier based on a small database (less than 100 images representing one class). The network training process can be shortened to a few minutes.

COVID-19 detection on Chest X-ray images: A comparison of CNN architectures and ensembles

COVID-19 quickly became a global pandemic after only four months of its first detection. It is crucial to detect this disease as soon as possible to decrease its spread. The use of chest X-ray (CXR) images became an effective screening strategy, complementary to the reverse transcription-polymerase chain reaction (RT-PCR). Convolutional neural networks (CNNs) are often used for automatic image classification and they can be very useful in CXR diagnostics. In this paper, 21 different CNN architectures are tested and compared in the task of identifying COVID-19 in CXR images. They were applied to the COVIDx8B dataset, a large COVID-19 dataset with 16,352 CXR images coming from patients of at least 51 countries. Ensembles of CNNs were also employed and they showed better efficacy than individual instances. The best individual CNN instance results were achieved by DenseNet169, with an accuracy of 98.15% and an F1 score of 98.12%. These were further increased to 99.25% and 99.24%, respectively, through an ensemble with five instances of DenseNet169. These results are higher than those obtained in recent works using the same dataset.

A Fast Inference Vision Transformer for Automatic Pavement Image Classification and Its Visual Interpretation Method

Traditional automatic pavement distress detection methods using convolutional neural networks (CNNs) require a great deal of time and resources for computing and are poor in terms of interpretability. Therefore, inspired by the successful application of Transformer architecture in natural language processing (NLP) tasks, a novel Transformer method called LeViT was introduced for automatic asphalt pavement image classification. LeViT consists of convolutional layers, transformer stages where Multi-layer Perception (MLP) and multi-head self-attention blocks alternate using the residual connection, and two classifier heads. To conduct the proposed methods, three different sources of pavement image datasets and pre-trained weights based on ImageNet were attained. The performance of the proposed model was compared with six state-of-the-art (SOTA) deep learning models. All of them were trained based on transfer learning strategy. Compared to the tested SOTA methods, LeViT has less than 1/8 of the parameters of the original Vision Transformer (ViT) and 1/2 of ResNet and InceptionNet. Experimental results show that after training for 100 epochs with a 16 batch-size, the proposed method acquired 91.56% accuracy, 91.72% precision, 91.56% recall, and 91.45% F1-score in the Chinese asphalt pavement dataset and 99.17% accuracy, 99.19% precision, 99.17% recall, and 99.17% F1-score in the German asphalt pavement dataset, which is the best performance among all the tested SOTA models. Moreover, it shows superiority in inference speed (86 ms/step), which is approximately 25% of the original ViT method and 80% of some prevailing CNN-based models, including DenseNet, VGG, and ResNet. Overall, the proposed method can achieve competitive performance with fewer computation costs. In addition, a visualization method combining Grad-CAM and Attention Rollout was proposed to analyze the classification results and explore what has been learned in every MLP and attention block of LeViT, which improved the interpretability of the proposed pavement image classification model.

Dermoscopic Image Classification with Neural Style Transfer

Skin cancer, the most commonly found human malignancy, is primarily diagnosed visually via dermoscopic analysis, biopsy and histopathological examination. However, unlike other types of cancer, automated image classification of skin lesions is deemed to be more challenging due to the irregularity and variability in the lesions’ appearances. In this work, we propose an adaptation of the Neural Style Transfer (NST) as a novel image pre-processing step for skin lesion classification problems. We represent each dermoscopic image as a style image and transfer the style of the lesion onto a homogeneous content image. This transfers the main variability of each lesion onto the same localized region, which allows us to integrate the generated images together and extract low-rank latent embeddings via tensor decomposition. We evaluated the performance of our model on competition data sets collected and pre-processed from the International Skin Imaging Collaboration (ISIC) database. We show that the classification performance based on the extracted tensor features using the style-transferred images significantly outperforms that of the raw images by more than 10%, and is also competitive with well-studied, pre-trained CNN models using transfer learning. Additionally, the tensor decomposition also affords clinical interpretations and insights by examining the images which correspond to the largest loadings in the top style embedding features as identified by the common supervised learning models.

Automatic Classification of Images with Skin Cancer Using Artificial Intelligence

Skin cancer is the most common malignant neoplasty in the world, it is a public health problem, which has increased in recent years due to environmental changes, different lifestyles, sun exposure, among others. One way to detect skin cancer is by analyzing medical images, analyzing these images can get the detection of any abnormalities. In this paper, several block programming models are implemented with classifiers for the recognition of medical images of skin cancer. Preprocessing, manipulation, and computer vision to extract the relevant characteristics of the images are the starting point for obtaining appropriate classification values. The main objective of this project is to perform the analysis of a set of classification techniques, as well as to verify that combination of image processing operations and classification tools provide better performance compared to the classification values of the original images. Images of three kinds of skin cancer type were used: melanocytic nevi, melanoma, and benign lesions similar to keratosis. Each category contains 200 images. The images were subjected to a set of filters, to later use different classification algorithms. 6 types of filters and 5 different classification techniques were implemented. The results obtained allow us to see the feasibility of the proposed method.

Diffeomorphic transforms for data augmentation of highly variable shape and texture objects

Background and objective: Training a deep convolutional neural network (CNN) for automatic image classification requires a large database with images of labeled samples. However, in some applications such as biology and medicine only a few experts can correctly categorize each sample. Experts are able to identify small changes in shape and texture which go unnoticed by untrained people, as well as distinguish between objects in the same class that present drastically different shapes and textures. This means that currently available databases are too small and not suitable to train deep learning models from scratch. To deal with this problem, data augmentation techniques are commonly used to increase the dataset size. However, typical data augmentation methods introduce artifacts or apply distortions to the original image, which instead of creating new realistic samples, obtain basic spatial variations of the original ones.Methods: We propose a novel data augmentation procedure which generates new realistic samples, by combining two samples that belong to the same class. Although the idea behind the method described in this paper is to mimic the variations that diatoms experience in different stages of their life cycle, it has also been demonstrated in glomeruli and pollen identification problems. This new data augmentation procedure is based on morphing and image registration methods that perform diffeomorphic transformations.Results: The proposed technique achieves an increase in accuracy over existing techniques of 0.47%, 1.47%, and 0.23% for diatom, glomeruli and pollen problems respectively.Conclusions: For the Diatom dataset, the method is able to simulate the shape changes in different diatom life cycle stages, and thus, images generated resemble newly acquired samples with intermediate shapes. In fact, the other methods compared obtained worse results than those which were not using data augmentation. For the Glomeruli dataset, the method is able to add new samples with different shapes and degrees of sclerosis (through different textures). This is the case where our proposed DA method is more beneficial, when objects highly differ in both shape and texture. Finally, for the Pollen dataset, since there are only small variations between samples in a few classes and this dataset has other features such as noise which are likely to benefit other existing DA techniques, the method still shows an improvement of the results.

Supplementing a Web-based Exposure Estimation System with Deep Learning for Automatic Classification of CT Images to Increase the Efficiency of Effective Dose Estimation

Supplementing a Web-based Exposure Estimation System with Deep Learning for Automatic Classification of CT Images to Increase the Efficiency of Effective Dose Estimation

WDCCNet: Weighted Double-Classifier Constraint Neural Network for Mammographic Image Classification.

The early detection and timely treatment of breast cancer can save lives. Mammography is one of the most efficient approaches to screening early breast cancer. An automatic mammographic image classification method could improve the work efficiency of radiologists. Current deep learning-based methods typically use the traditional softmax loss to optimize the feature extraction part, which aims to learn the features of mammographic images. However, previous studies have shown that the feature extraction part cannot learn discriminative features from complex data using the standard softmax loss. In this paper, we design a new architecture and propose respective loss functions. Specifically, we develop a double-classifier network architecture that constrains the extracted features' distribution by changing the classifiers' decision boundaries. Then, we propose the double-classifier constraint loss function to constrain the decision boundaries so that the feature extraction part can learn discriminative features. Furthermore, by taking advantage of the architecture of two classifiers, the neural network can detect the difficult-to-classify samples. We propose a weighted double-classifier constraint method to make the feature extract part pay more attention to learning difficult-to-classify samples' features. Our proposed method can be easily applied to an existing convolutional neural network to improve mammographic image classification performance. We conducted extensive experiments to evaluate our methods on three public benchmark mammographic image datasets. The results showed that our methods outperformed many other similar methods and state-of-the-art methods on the three public medical benchmarks. Our code and weights can be found on GitHub.

HER2-ResNet: A HER2 classification method based on deep residual network.

BACKGROUND: HER2 gene expression is one of the main reference indicators for breast cancer detection and treatment, and it is also an important target for tumor targeted therapy drug selection. Therefore, the correct detection and evaluation of HER2 gene expression has important value for clinical treatment of breast cancer.OBJECTIVE: The study goal is to better classify HER2 images.METHODS: For general convolution neural network, with the increase of network layers, over fitting phenomenon is often very serious, which requires setting the value of random descent ratio, and parameter adjustment is often time-consuming and laborious, so this paper uses residual network, with the increase of network layer, the accuracy will not be reduced.RESULTS: In this paper, a HER2 image classification algorithm based on improved residual network is proposed. Experimental results show that the proposed HER2 network has high accuracy in breast cancer assessment.Conclusion: Taking HER2 images in Stanford University database as experimental data, the accuracy of HER2 image automatic classification is improved through experiments. This method will help to reduce the detection intensity and improve the accuracy of HER2 image classification.

Genetic programming for automatic skin cancer image classification

Developing a computer-aided diagnostic system for detecting various types of skin malignancies from images has attracted many researchers. However, analyzing the behaviors of algorithms is as important as developing new systems in order to establish the effectiveness of a system in real-time situations which impacts greatly how well it can assist the dermatologist in making a diagnosis. Unlike many machine learning approaches such as Artificial Neural Networks, Genetic Programming (GP) automatically evolves models with its dynamic representation and flexibility. This study aims at analyzing recently developed GP-based approaches to skin image classification. These approaches have utilized the intrinsic feature selection and feature construction ability of GP to effectively construct informative features from a variety of pre-extracted features. These features encompass local, global, texture, color and multi-scale image properties of skin images. The performance of these GP methods is assessed using two real-world skin image datasets captured from standard camera and specialized instruments, and compared with six commonly used classification algorithms as well as existing GP methods. The results reveal that these constructed features greatly help improve the performance of the machine learning classification algorithms. Unlike “black-box” algorithms like deep neural networks, GP models are interpretable, therefore, our analysis shows that these methods can help dermatologists identify prominent skin image features. Further, it can help researchers identify suitable feature extraction methods for images captured from a specific instrument. Being fast, these methods can be deployed for making a quick and effective diagnosis in actual clinic situations.

Image feature extraction and galaxy classification: a novel and efficient approach with automated machine learning

ABSTRACT In this work, we explore the possibility of applying machine learning methods designed for 1D problems to the task of galaxy image classification. The algorithms used for image classification typically rely on multiple costly steps, such as the point spread function deconvolution and the training and application of complex Convolutional Neural Networks of thousands or even millions of parameters. In our approach, we extract features from the galaxy images by analysing the elliptical isophotes in their light distribution and collect the information in a sequence. The sequences obtained with this method present definite features allowing a direct distinction between galaxy types. Then, we train and classify the sequences with machine learning algorithms, designed through the platform Modulos AutoML. As a demonstration of this method, we use the second public release of the Dark Energy Survey (DES DR2). We show that we are able to successfully distinguish between early-type and late-type galaxies, for images with signal-to-noise ratio greater than 300. This yields an accuracy of $86{{\ \rm per\ cent}}$ for the early-type galaxies and $93{{\ \rm per\ cent}}$ for the late-type galaxies, which is on par with most contemporary automated image classification approaches. The data dimensionality reduction of our novel method implies a significant lowering in computational cost of classification. In the perspective of future data sets obtained with e.g. Euclid and the Vera Rubin Observatory, this work represents a path towards using a well-tested and widely used platform from industry in efficiently tackling galaxy classification problems at the peta-byte scale.

Automated characterisation of microglia in ageing mice using image processing and supervised machine learning algorithms

The resident macrophages of the central nervous system, microglia, are becoming increasingly implicated as active participants in neuropathology and ageing. Their diverse and changeable morphology is tightly linked with functions they perform, enabling assessment of their activity through image analysis. To better understand the contributions of microglia in health, senescence, and disease, it is necessary to measure morphology with both speed and reliability. A machine learning approach was developed to facilitate automatic classification of images of retinal microglial cells as one of five morphotypes, using a support vector machine (SVM). The area under the receiver operating characteristic curve for this SVM was between 0.99 and 1, indicating strong performance. The densities of the different microglial morphologies were automatically assessed (using the SVM) within wholemount retinal images. Retinas used in the study were sourced from 28 healthy C57/BL6 mice split over three age points (2, 6, and 28-months). The prevalence of ‘activated’ microglial morphology was significantly higher at 6- and 28-months compared to 2-months (p < .05 and p < .01 respectively), and ‘rod’ significantly higher at 6-months than 28-months (p < 0.01). The results of the present study propose a robust cell classification SVM, and further evidence of the dynamic role microglia play in ageing.

WBC-AMNet: Automatic classification of WBC images using deep feature fusion network based on focalized attention mechanism.

The recognition and classification of White Blood Cell (WBC) play a remarkable role in blood-related diseases (i.e., leukemia, infections) diagnosis. For the highly similar morphology of different WBC subtypes, it is too confused to classify the WBC effectively and accurately for visual observation of blood cell smears. This paper proposes a Deep Convolutional Neural Network (DCNN) with feature fusion strategies, named WBC-AMNet, for automatically classifying WBC subtypes based on focalized attention mechanism. To obtain more localized attention of CNN, the fusion features of the first and the last convolutional layer are extracted by focalized attention mechanism combining Squeeze-and-Excitation (SE) and Gather-Excite (GE) modules. The new method performs successfully in classifying monocytes, neutrophils, lymphocytes, and eosinophils on the complex background with an overall accuracy of 95.66%, better than that of general CNNs. The multi-classification accuracy of WBC-AMNet with the background segmentation is over 98% in all cases. In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) is employed to visualize the attention heatmaps of different feature maps.

Classification of histopathological images of breast cancer using an improved convolutional neural network model.

The accurately automatic classification of medical pathological images has always been an important problem in the field of deep learning. However, the traditional manual extraction of features and image classification usually requires in-depth knowledge and more professional researchers to extract and calculate high-quality image features. This kind of operation generally takes a lot of time and the classification effect is not ideal. In order to solve these problems, this study proposes and tests an improved network model DenseNet-201-MSD to accomplish the task of classification of medical pathological images of breast cancer. First, the image is preprocessed, and the traditional pooling layer is replaced by multiple scaling decomposition to prevent overfitting due to the large dimension of the image data set. Second, the BN algorithm is added before the activation function Softmax and Adam is used in the optimizer to optimize performance of the network model and improve image recognition accuracy of the network model. By verifying the performance of the model using the BreakHis dataset, the new deep learning model yields image classification accuracy of 99.4%, 98.8%, 98.2%and 99.4%when applying to four different magnifications of pathological images, respectively. The study results demonstrate that this new classification method and deep learning model can effectively improve accuracy of pathological image classification, which indicates its potential value in future clinical application.

Automatic Classification of Diabetic Foot Ulcer Images - A Transfer-Learning Approach to Detect Wound Maceration.

Diabetic foot ulcer (DFU) is a chronic wound and a common diabetic complication as 2% - 6% of diabetic patients witness the onset thereof. The DFU can lead to severe health threats such as infection and lower leg amputations, Coordination of interdisciplinary wound care requires well-written but time-consuming wound documentation. Artificial intelligence (AI) systems lend themselves to be tested to extract information from wound images, e.g. maceration, to fill the wound documentation. A convolutional neural network was therefore trained on 326 augmented DFU images to distinguish macerated from unmacerated wounds. The system was validated on 108 unaugmented images. The classification system achieved a recall of 0.69 and a precision of 0.67. The overall accuracy was 0.69. The results show that AI systems can classify DFU images for macerations and that those systems could support clinicians with data entry. However, the validation statistics should be further improved for use in real clinical settings. In summary, this paper can contribute to the development of methods to automatic wound documentation.

Auroral Image Classification With Very Limited Labeled Data Using Few-Shot Learning

Auroral image classification is an important task for polar research. At present, the auroral image classification mainly focuses on common auroral types with a lot of labeled samples, where the classification performance is heavily dependent on large labeled datasets for supervised training. However, in many instances, such as for rare auroras or newly discovered auroras, it is not feasible to obtain ample category labels. This letter proposes a few-shot learning-based solution for auroral image classification with very limited labeled samples. Since auroral data are always collected by ground-based all-sky imagers with little domain differences, we incorporate prior knowledge into auroral image representation with limited labeled samples using the labels from previous auroral image classification studies. Also, the cosine classifier is employed to avoid the overfitting problem caused by the very few labels of the auroral images to be classified. When providing one and five labeled samples for each category, our solution achieves an average classification accuracy of 87.14% and 95.33% on the novel auroral category dataset, respectively. The effectiveness of this solution is further demonstrated by auroral image retrieval and temporal occurrence distributions. The experimental results show that the proposed solution provides a new idea for the automatic classification and retrieval of auroral images, especially for those cases where there are only very few labeled samples.

A Novel Optimizer in Deep Neural Network for Diabetic Retinopathy Classification

In severe cases, diabetic retinopathy can lead to blindness. For decades, automatic classification of diabetic retinopathy images has been a challenge. Medical image processing has benefited from advances in deep learning systems. To enhance the accuracy of image classification driven by Convolutional Neural Network (CNN), balanced dataset is generated by data augmentation method followed by an optimized algorithm. Deep neural networks (DNN) are frequently optimized using gradient (GD) based techniques. Vanishing gradient is the main drawback of GD algorithms. In this paper, we suggest an innovative algorithm, to solve the above problem, Hypergradient Descent learning rate based Quasi hyperbolic (HDQH) gradient descent to optimize the weights and biases. The algorithms only use first order gradients, which reduces computation time and storage space requirements. The algorithms do not require more tuning of the learning rates as the learning rate tunes itself by means of gradients. We present empirical evaluation of our algorithm on two public retinal image datasets such as Messidor and DDR by using Resnet18 and Inception V3 architectures. The findings of the experiment show that the efficiency and accuracy of our algorithm outperforms the other cutting-edge algorithms. HDQHAdam shows the highest accuracy of 97.5 on Resnet18 and 95.7 on Inception V3 models respectively.

A Novel Vision Transformer Approach for Automatic Asphalt Pavement Image Classification and its Visual Interpretation Methods

A Novel Vision Transformer Approach for Automatic Asphalt Pavement Image Classification and its Visual Interpretation Methods

ResNet50-Based Effective Model for Breast Cancer Classification Using Histopathology Images

Breast cancer is considered an immense threat and one of the leading causes of mortality in females. It is curable only when detected at an early stage. A standard cancer diagnosis approach involves detection of cancer-related anomalies in tumour histopathology images. Detection depends on the accurate identification of the landmarks in the visual artefacts present in the slide images. Researchers are continuously striving to develop automatic machine-learning algorithms for processing medical images to assist in tumour detection. Nowadays, computer-based automated systems play an important role in cancer image analysis and help healthcare experts make rapid and correct inferences about the type of cancer. This study proposes an effective convolutional neural network-based (CNN-based) model that exploits the transfer-learning technique for automatic image classification between malignant and benign tumour, using histopathology images. Resnet50 architecture has been trained on new dataset for feature extraction, and fully connected layers have been fine-tuned for achieving highest training, validation and test accuracies. The result illustrated state-of-the-art performance of the proposed model with highest training, validation and test accuracies as 99.70%, 99.24% and 99.24%, respectively. Classification accuracy is increased by 0.66% and 0.2% when compared with similar recent studies on training and test data results. Average precision and F1 score have also improved, and receiver operating characteristic (RoC) area has been achieved to 99.1%. Thus, a reliable, accurate and consistent CNN model based on pre-built Resnet50 architecture has been developed.

Deep Learning Image Analysis of High-Throughput Toxicology Assay Images.

High-throughput chemical screening approaches often employ microscopy to capture photomicrographs from multi-well cell culture plates, generating thousands of images that require time-consuming human analysis. To automate this subjective and time-consuming manual process, we have developed a method that uses deep learning to automatically classify digital assay images. We have trained a convolutional neural network (CNN) to perform binary and multi-class classification. The binary classifier binned assay images into healthy (comparable to untreated controls) and altered (not comparable to untreated-control) classes with >98% accuracy; the multi-class classifier assigned "Healthy," "Intermediate" and "Altered" labels to assay images with >95% accuracy. Our dataset comprised high-resolution assay images from primary human hepatocytes and undifferentiated (proliferating) and differentiated 2D cultures of HepaRG cells. In this study we have focused on testing and fine-tuning various CNN architectures, including ResNet 34, 50 and 101. To visualize regions in the images that the CNN model used for classification, we employed Class Activation Maps (CAM). This allowed us to better understand the inner workings of the neural network and led to additional optimizations of the algorithm. The results indicate a strong correspondence between dosage and classifier-predicted scores, suggesting that these scores might be useful in further characterizing benchmark dose. Together, these results clearly demonstrate that deep-learning based automated image classification of cell morphology changes upon chemical-induced stress can yield highly accurate and reproducible assessments of cytotoxicity across a variety of cell types.