Transfer Learning For Image Classification Research Articles

Ensemble Transfer Learning for Image Classification

The deep learning (DL) techniques used for image classification might not deliver the desired level of classification accuracy as some features belonging to some class of a dataset are missed during feature extraction. The ensemble learning (EL) based model improves classification accuracy by combining the strengths of individual classifiers. As a result, those features that were missed during feature extraction by a specific DL technique will be taken care of by another DL technique in an ensemble DL approach. In this paper, averaging EL (AENet), weighted averaging EL (WAENet), and stacking EL (StackedNet) approaches are proposed, considering the DenseNet201, EcientNetB0, and ResNetRS101 as base models. The predictions of the base models are averaged to generate the AENet. The WAENet is constructed by assigning weights to each base model based on their prediction and then taking their average. Similarly, the Stacked-Net is developed by considering the DenseNet201, EcientNetB0, and ResNetRS101 as base-learners and ResNetRS101 as meta-learner. Analysed performance of the considered pre-trained base models and the developed EL models on the standard and application-specific datasets such as MiniImageNet, CIFAR10, CIFAR100, Plant Village (PV), Tomato, Covid-19 and 9IndianFood. 80% of the datasets were used to train and 20% to test the base and proposed models. The models are trained for an epoch of 30, considering a learning rate of 0.001 and adam optimizer. The stackedNet delivered better results than others.

Transfer learning for acoustic cement bond evaluation: An image classification approach using acoustic variable Density log

Ensuring the integrity of wellbore cement is critical for the environmental protection, safety, and economic viability of oil, gas, geothermal, and Carbon Capture and Storage (CCS) operations. Acoustic logging is the most commonly used evaluation method, and interpreting its data requires significant expertise, often informs high-stakes decisions. Machine Learning (ML) has been used to help the data interpretation, but they are heavily reliant on meticulous feature engineering—a labor-intensive and expertise-driven task. Moreover, ML methods cannot tell which portions of input acoustic data are most important in the cement evaluation task. This level of analysis is vital, as it can significantly inform and improve future strategies for acoustic logging data collection and signal processing, ultimately improving the cement evaluation results. To address these challenges, we present a generalized workflow. First, we convert Variable Density Log (VDL) data into images using the Continuous Wavelet Transform (CWT). Then, we apply transfer learning for image classification to enhance the classification of wellbore cement isolation. In transfer learning stage, the images are processed using several pre-trained image classifiers—Xception, VGG16, MobileNetV2, and ResNet50—originally trained on the extensive ImageNet database containing over 14 million images. To adapt these models for our specific task, we added a global average pooling layer to reduce feature map dimensionality, followed by a fully connected layer with a Rectified Linear Unit (ReLU) activation function to introduce non-linearity and enhance learning capability. We replaced the original final layer of each classifier with a new three-neuron layer using softmax activation, tailored for multi-class classification. We preserved the general characteristics learned from the ImageNet dataset by freezing all original layers except for the newly added ones and compiled the models using the Adam optimizer. The performance of these adapted models was evaluated based on accuracy, loss, and F1 score. We applied this workflow to two distinct datasets: the first from a Norwegian wellbore and the second from a hydrocarbon well in China. In the Norwegian case, the Xception model achieved a classification accuracy of 97.4%, which was further refined to 99.0% through the implementation of Gradient-weighted Class Activation Mapping (Grad-CAM), revealing critical frequency ranges that traditional ML methods could not identify. In the Chinese case, employing the same models and workflow, the MobileNetV2 algorithm demonstrated exceptional performance, achieving a 99.6% accuracy in predicting cement isolation. In addition to achieving high accuracy, our results reveal that the VDL frequency content between 20 and 25 kHz is more critical for cement evaluation in the investigated wells than other frequency ranges, a level of analysis unattainable with traditional ML methods. Furthermore, our approach eliminates the need for deep knowledge in feature engineering, typically requisite in traditional ML, and operates effectively without any feature engineering. These results underscore the efficacy of our approach across diverse geological settings and highlight the potential of transfer learning to revolutionize cement evaluation automation, inspiring its broader application in the industry.

Deep Transfer Learning for Image Classification of Phosphorus Nutrition States in Individual Maize Leaves

Computer vision is a powerful technology that has enabled solutions in various fields by analyzing visual attributes of images. One field that has taken advantage of computer vision is agricultural automation, which promotes high-quality crop production. The nutritional status of a crop is a crucial factor for determining its productivity. This status is mediated by approximately 14 chemical elements acquired by the plant, and their determination plays a pivotal role in farm management. To address the timely identification of nutritional disorders, this study focuses on the classification of three levels of phosphorus deficiencies through individual leaf analysis. The methodological steps include: (1) using different capture devices to generate a database of images composed of laboratory-grown maize plants that were induced to either total phosphorus deficiency, medium deficiency, or total nutrition; (2) processing the images with state-of-the-art transfer learning architectures (i.e., VGG16, ResNet50, GoogLeNet, DenseNet201, and MobileNetV2); and (3) evaluating the classification performance of the models using the created database. The results show that the DenseNet201 model achieves superior performance, with 96% classification accuracy. However, the other studied architectures also demonstrate competitive performance and are considered state-of-the-art automatic leaf nutrition deficiency detection tools. The proposed method can be a starting point to fine-tune machine-vision-based solutions tailored for real-time monitoring of crop nutritional status.

Quantum transfer learning for image classification

Quantum machine learning, an important element of quantum computing, recently has gained research attention around the world. In this paper, we have proposed a quantum machine learning model to classify images using a quantum classifier. We exhibit the results of a comprehensive quantum classifier with transfer learning applied to image datasets in particular. The work uses hybrid transfer learning technique along with the classical pre-trained network and variational quantum circuits as their final layers on a small scale of dataset. The implementation is carried out in a quantum processor of a chosen set of highly informative functions using PennyLane a cross-platform software package for using quantum computers to evaluate the high-resolution image classifier. The performance of the model proved to be more accurate than its counterpart and outperforms all other existing classical models in terms of time and competence.

Design and development of smart Internet of Things-based solid waste management system using computer vision.

Municipal solid waste (MSW) management currently requires critical attention in ensuring the best principles of socio-economic attributes such as environmental protection, economic sustainability, and mitigation of human health problems. Numerous surveys on the waste management system reveal that approximately 90% of the MSW systems are improperly disposing the wastages in open dumps and landfills. Classifying the wastages into biodegradable and non-biodegradable helps converting them into usable energy and disposing properly. The advancements of effective computational approaches like artificial intelligence and image processing provide wide range of solutions for the present problem identified in MSW management. The computational approaches can be programmed to classify wastes that help to convert them into usable energy. Existing methods of waste classification in MSW remain unresolved due to poor accuracy and higher error rate. This paper presents an experimented effective computer vision–based MSW management solution with the help of the Internet of Things (IoT), and machine learning (ML) techniques namely regression, classification, clustering, and correlation rules for the perception of solid waste images. A ground-up built convolutional neural network (CNN) and CNN by the inception of ResNet V2 models trained through transfer learning for image classification. ResNet V2 supports training large datasets in deep neural networks to achieve improved accuracy and reduced error rate in identity mapping. In addition, batch normalization and mixed hybrid pooling techniques are incorporated in CNN to improve stability and yield state of art performance. The proposed model identifies the type of waste and classifies them as biodegradable or non-biodegradable to collect in respective waste bins precisely. Furthermore, observation of performance metrics, accuracy, and loss ensures the effective functions of the proposed model compared to other existing models. The proposed ResNet-based CNN performs waste classification with 19.08% higher accuracy and 34.97% lower loss than the performance metrics of other existing models.

Classical–Quantum Transfer Learning for Image Classification

Classical–Quantum Transfer Learning for Image Classification

Transfer learning for image classification using VGG19: Caltech-101 image data set.

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.

Activation ensemble generative adversarial network transfer learning for image classification

Transfer learning provides a useful solution to learn a new conceptual domain from few examples, which exploits prior knowledge from a related domain. We proposed a simple and yet effective transfer learning method for image classification that constructs an activation ensemble generative adversarial net (AE-GAN) to transfer knowledge from one dataset to another. The AE-GAN is mainly composed of three convolutional layers and adopts an ensemble of multiple activation functions. Experimental results on five benchmark datasets show that when only a few samples are available for training a target task, leveraging datasets from other related datasets by AE-GAN can significantly improve the performance for image classification with a small set of samples.

Investigation of transfer learning for image classification and impact on training sample size

Recent developments in deep learning have brought huge breakthroughs in the image processing area, which triggered numerous successful applications and positively impacted the current big data context of Industry 4.0. On the other hand, it is widely known that large amounts of training data are required to train a deep learning model with millions of parameters, which limits its application in many industrial applications where sufficient data resources are lacking. Transfer learning is one of the practical solutions to reduce the data required for training, which tries to reuse learned knowledge for similar tasks. Nevertheless, many technical details of transfer learning implementation are not well documented. Therefore, in this work, two datasets collected from plastics manufacturing processes were studied to investigate different transfer learning approaches and implementation details for high-performance model building under the constraint of limited available training data. Different transfer learning implementations are compared and important technical details are also discussed in this study. Through this study, the minimum number of training samples can be estimated. Transfer learning is compared with the newly developed few-shot learning approach as a brief comparative study. Finally, this work summarizes practical guidelines for the development of image classification models with limited data resources.

Domain class consistency based transfer learning for image classification across domains

Distribution mismatch between the modeling data and the query data is a known domain adaptation issue in machine learning. To this end, in this paper, we propose a l2,1-norm based discriminative robust kernel transfer learning (DKTL) method for high-level recognition tasks. The key idea is to realize robust domain transfer by simultaneously integrating domain-class-consistency (DCC) metric based discriminative subspace learning, kernel learning in reproduced kernel Hilbert space, and representation learning between source and target domain. The DCC metric includes two properties: domain-consistency used to measure the between-domain distribution discrepancy and class-consistency used to measure the within-domain class separability. The essential objective of the proposed transfer learning method is to maximize the DCC metric, which is equivalently to minimize the domain-class-inconsistency (DCIC), such that domain distribution mismatch and class inseparability are well formulated and unified simultaneously. The merits of the proposed method include (1) the robust sparse coding selects a few valuable source data with noises (outliers) removed during knowledge transfer, and (2) the proposed DCC metric can pursue more discriminative subspaces of different domains. As a result, the maximum class-separability is also well guaranteed. Extensive experiments on a number of visual datasets demonstrate the superiority of the proposed method over other state-of-the-art domain adaptation and transfer learning methods.

Correction: Class Prediction and Feature Selection with Linear Optimization for Metagenomic Count Data

A reader recently pointed out certain similarities between the text in the article and that from some previous publications. Although our method in PLOS ONE was novel and citations to many of the relevant publications were included, the text should not have been used verbatim. The authors would like to thank the reader and apologize for the overlap in text. The overlap in the text relates to the Introduction, Methods and Discussion sections of the article, where sentences from previous publications were reproduced, this relates to the following fragments in the text: In the Methods section of the article there is some overlap in text with the first author Zhenqiu Liu’s previous publication, cited as reference 32: Liu Z, Hsiao W, Cantarel B, Drbek E, Fraser-Liggett C (2011). Sparse distance based learning for simultaneous multiclass classification and feature selection of metagenomic data. Bioinformatics 27(23): 3242–3249. The study reported in Bioinformatics and that published in PLOS ONE studied the same multiclass classification problem for metagenomic count data, and thus mathematical notations and the problem are defined in a similar way in the two articles. In addition, the description of the data normalization procedures in the PLOS ONE article is given in a way similar to that in the earlier publication which in Bioinformatics. The description of the methods that overlap with that in the publication in Bioinformatics relates to the first paragraphs of the Methods section before the section titled and Penalized SVM Methods. With regard to equations (6) and (7), our article cited the following article by Quattroni et al. as reference 35: Quattoni A, Carreras X, Collins M, and Darrell T (2009). An Efficient Projection for L1,∞ Regularization, ICML 2009 However we acknowledge that it would have been more appropriate to cite this earlier work by the same group: Quattoni A, Collins M, and Darrell T (2008), Transfer Learning for Image Classification with Sparse Prototype Representations. In Proceedings of CVPR 2008. Other instances of text overlap occur in the Introduction and in the Discussion sections of the article. In the first paragraph of the Introduction, there are sentences that overlap with text in references 4 and 7, as well as Schloss et al. BMC Bioinformatics. 2008 Jan 23;9:34. doi: 10.1186/1471-2105-9-34, which was not cited in the article: ‘The majority of microbes reside in the gut, have a profound influence on human physiology and nutrition, and are crucial for human life. Metagenomics, the culture-independent isolation and characterization of DNA from uncultured microorganisms, has facilitated the analysis of the functional biodiversity harbored in the large reservoir of uncultured bacteria and archaea.’ ‘Recent advances in genome sequencing technologies have made obtaining a complete metagenomic sequencing more tractable. Having on hand such a large number of microbial genomes has changed the nature of microbiology and of microbial evolution studies. By providing the ability to examine the relationship of genome structure and function across many different species, these data have also opened up the fields of comparative genomics and of systems biology. A main promise of metagenomics is that it will accelerate drug discovery and biotechnology by providing new genes with novel functions’ In the Results section, there are sentences that overlap with text from references 26 and 38 in the article: ‘ are simulated from the Gamma distribution with a mean () of 100 and variance () of 1000. We simulated 1000 features for each sample from NB distributions, which contained the first 5 relevant features having different distributions with distinguished s. We used two-fold cross validation to evaluate the method. First, we normalized the data with proportion and arcsin transformations, and then divided the data into training and test equal subsets. The training subset was used for model construction, while the test subset was used to evaluate performance. The model parameters are determined from only the training data with leave-one-out cross-validation.’ ‘Bacteria thrive on and within the human body. One of the largest human-associated microbial habitats is the skin surface, which harbors large numbers of bacteria that can have important effects on health.’ Even though we cited the articles above, we apologize that the information was not rewritten more carefully. The issue of existing text overlap has no bearing on the results and conclusions of the study.

Heterogeneous Transfer Learning for Image Classification

Transfer learning as a new machine learning paradigm has gained increasing attention lately. In situations where the training data in a target domain are not sufficient to learn predictive models effectively, transfer learning leverages auxiliary source data from other related source domains for learning. While most of the existing works in this area only focused on using the source data with the same structure as the target data, in this paper, we push this boundary further by proposing a heterogeneous transfer learning framework for knowledge transfer between text and images. We observe that for a target-domain classification problem, some annotated images can be found on many social Web sites, which can serve as a bridge to transfer knowledge from the abundant text documents available over the Web. A key question is how to effectively transfer the knowledge in the source data even though the text can be arbitrarily found. Our solution is to enrich the representation of the target images with semantic concepts extracted from the auxiliary source data through a novel matrix factorization method. By using the latent semantic features generated by the auxiliary data, we are able to build a better integrated image classifier. We empirically demonstrate the effectiveness of our algorithm on the Caltech-256 image dataset.