Minor Classes Research Articles

Active Broad-Transfer Learning Algorithm for Class-Imbalanced Fault Diagnosis

Knowledge transfer with class-imbalanced data is a challenge in predictive maintenance and fault diagnosis. Deep learning algorithms have provided promising results in fault diagnosis. However, their prediction performance is affected by class-imbalanced data in cross-domain tasks. Broad learning algorithms present promising performance in handling class-imbalanced domain-adaptation (CIDA) problems. In the presence of a domain shift, active broad transfer for class-imbalanced learning (ABTCI), an active broad-transfer learning algorithm for CIDA, is proposed. First, the ABTCI algorithm extracts the time–frequency features and feeds them into a recurrent cell to capture spatial–temporal features. Subsequently, it augments the feature space using a sparse autoencoder and an orthogonal mapping projector. By solving the ridge regression problem, the classifier is initialized. Next, the algorithm samples the target data with reliable pseudo-labels and synthesizes new data using random intraclass interpolation among the minor classes containing source and target knowledge. Finally, the classifier is updated using an incremental continuous learning strategy. The performance of the ABTCI algorithm is validated using three datasets, which include 20 class-balanced and 27 class-imbalanced transfer tasks. The performance of the proposed algorithm benchmarked against other deep-learning algorithms is promising.

Phytochemical Composition, Antioxidant and Antiproliferative Activities of Citrus hystrix, Citrus limon, Citrus pyriformis, and Citrus microcarpa Leaf Essential Oils against Human Cervical Cancer Cell Line

The essential oil derived from Citrus plants has long been used for medicinal purposes, due to its broad spectrum of therapeutic characteristics. To date, approximately 162 Citrus species have been identified, and many investigational studies have been conducted to explore the pharmacological potential of Citrus spp. oils. This study investigated the volatile constituents of essential oil distilled from the leaves of C. hystrix, C. limon, C. pyriformis, and C. microcarpa, using gas chromatography-quadrupole mass spectrometry. A total of 80 secondary compounds were tentatively identified, representing 84.88-97.99% of the total ion count and mainly comprising monoterpene (5.20-76.15%) and sesquiterpene (1.36-27.14%) hydrocarbons, oxygenated monoterpenes (3.91-89.52%) and sesquiterpenes (0.21-38.87%), and other minor chemical classes (0.10-0.52%). In particular, 27 compounds (1.19-39.06%) were detected across all Citrus species. Principal component analysis of the identified phytoconstituents and their relative quantities enabled differentiation of the Citrus leaf oils according to their species, with the loading variables contributing to these metabolic differences being identified. The Citrus leaf oils were tested for their antioxidant and antiproliferative activities using 2,2-diphenyl-1-picryl-hydrazylhydrate (DPPH) and 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assays. The results indicated that C. limon displayed the highest DPPH radical scavenging ability (IC50 value of 29.14 ± 1.97 mg/mL), while C. hystrix exhibited the lowest activity (IC50 value of 279.03 ± 10.37 mg/mL). On the other hand, all the Citrus oils exhibit potent antiproliferative activities against the HeLa cervical cancer cell line, with IC50 values of 11.66 μg/mL (C. limon), 20.41 μg/mL (C. microcarpa), 25.91 μg/mL (C. hystrix), and 87.17 μg/mL (C. pyriformis).

Activity of GABA neurons in the zona incerta and ventral lateral periaqueductal grey is biased towards sleep.

As in various brain regions the activity of gamma-aminobutyric acid (GABA) neurons is largely unknown, we measured in vivo changes in calcium fluorescence in GABA neurons in the zona incerta (ZI) and the ventral lateral periaqueductal grey (vlPAG), two areas that have been implicated in regulating sleep. vGAT-Cre mice were implanted with sleep electrodes, microinjected with rAAV-DIO-GCaMP6 into the ZI (n = 6) or vlPAG (n = 5) (isoflurane anesthesia) and a GRIN (Gradient-Index) lens inserted atop the injection site. Twenty-one days later, fluorescence in individual vGAT neurons was recorded over multiple REM cycles. Regions of interest corresponding to individual vGAT somata were automatically extracted with PCA-ICA analysis. In the ZI, 372 neurons were identified. Previously, we had recorded the activity of 310 vGAT neurons in the ZI and we combined the published dataset with the new dataset to create a comprehensive dataset of ZI vGAT neurons (total neurons = 682; mice = 11). In the vlPAG, 169 neurons (mice = 5) were identified. In both regions, most neurons were maximally active in REM sleep (R-Max; ZI = 51.0%, vlPAG = 60.9%). The second most abundant group was W-Max (ZI = 23.9%, vlPAG = 25.4%). In the ZI, but not in vlPAG, there were neurons that were NREMS-Max (11.7%). vlPAG had REMS-Off neurons (8.3%). In both areas, there were two minor classes: wake/REMS-Max and state indifferent. In the ZI, the NREMS-Max neurons fluoresced 30 s ahead of sleep onset. These descriptive data show that the activity of GABA neurons is biased in favor of sleep in two brain regions implicated in sleep.

Grid-Based Essential Urban Land Use Classification: A Data and Model Driven Mapping Framework in Xiamen City

Accurate and timely mapping of essential urban land use categories (EULUC) is vital to understanding urban land use distribution, pattern, and composition. Recent advances in leveraging big open data and machine learning algorithms have demonstrated the possibility of large-scale mapping of EULUC in a new cost-effective way. However, they are still limited by the transferability of samples, models, and classification results across space, particularly across different cities. Given the heterogeneities of environmental and socioeconomic conditions among cities, in-depth studies of data and model adaptation towards city-specific EULUC mappings are highly required to support policy making, and urban renewal planning and management practices. In addition, the trending need for timely and detailed small land unit data processing with finer data granularity becomes increasingly important. We proposed a City Meta Unit (CMU) data model and classification framework driven by multisource data and artificial intelligence (AI) algorithms to address these challenges. The CMU Framework was innovatively applied to systematically set up a grid-based data model and classify urban land use with an improved AI algorithm by applying Moore neighborhood correlations. Specifically, we selected Xiamen, Fujian, in China, a coastal city, as the typical testbed to implement this proposed framework and apply an AI transfer learning technique for grid and parcel land-use study. Experimental results with our proposed CMU framework showed that the grid-based land use classification performance achieves overall accuracies of 81.17% and 76.55% for level I (major classes) and level II (minor classes), which is much higher than the parcel-based land use classification (overall accuracies of 72.37% for level I, and 68.99% for level II). We further investigated the relationship between training sample size and classification performance and quantified the contribution of different data sources to urban land use classifications. The CMU framework makes data collections and processing intelligent and efficient, with finer granularity, saving time and cost by using existing open social data. Incorporating the CMU framework with the proposed grid-based model is an effective and new approach for urban land use classification, which can be flexibly extended and applied to various cities.

The Geographical Location as a Limitation for Starting Entrepreneurial Initiatives and Career Development

From the perspective of entrepreneurship, the geographical location is one of the important barriers/prerequisites for business development. Its role and place in the development of family entrepreneurship and the career development of young people has been relatively poorly studied. The aim of this paper is to provide an emperical research on the influence of the geographical location for starting entrepreneurial initiatives and career development. The object of this research are students having major or minor classes in the field of entrepreneurship from Albania, Bulgaria, Poland, Romania, Russia and Serbia. The subject of this research is on their attitude for career development in the field of family entrepreneurship and their geographical location. A standartised printed questionnaire has been used with Lickert scale for collecting the answers, translated into the national language of the respondents. The results are given in tables to compare the student intentions and outline their similarities. This paper could help the involved twelve universities to improve their teaching methods in a way their students to perceive the advantages of the geographical location and become part of the the regional entrepreneurs and companies.

SVD-CLAHE boosting and balanced loss function for Covid-19 detection from an imbalanced Chest X-Ray dataset

Covid-19 disease has had a disastrous effect on the health of the global population, for the last two years. Automatic early detection of Covid-19 disease from Chest X-Ray (CXR) images is a very crucial step for human survival against Covid-19. In this paper, we propose a novel data-augmentation technique, called SVD-CLAHE Boosting and a novel loss function Balanced Weighted Categorical Cross Entropy (BWCCE), in order to detect Covid 19 disease efficiently from a highly class-imbalanced Chest X-Ray image dataset. Our proposed SVD-CLAHE Boosting method is comprised of both oversampling and under-sampling methods. First, a novel Singular Value Decomposition (SVD) based contrast enhancement and Contrast Limited Adaptive Histogram Equalization (CLAHE) methods are employed for oversampling the data in minor classes. Simultaneously, a Random Under Sampling (RUS) method is incorporated in major classes, so that the number of images per class will be more balanced. Thereafter, Balanced Weighted Categorical Cross Entropy (BWCCE) loss function is proposed in order to further reduce small class imbalance after SVD-CLAHE Boosting. Experimental results reveal that ResNet-50 model on the augmented dataset (by SVD-CLAHE Boosting), along with BWCCE loss function, achieved 95% F1 score, 94% accuracy, 95% recall, 96% precision and 96% AUC, which is far better than the results by other conventional Convolutional Neural Network (CNN) models like InceptionV3, DenseNet-121, Xception etc. as well as other existing models like Covid-Lite and Covid-Net. Hence, our proposed framework outperforms other existing methods for Covid-19 detection. Furthermore, the same experiment is conducted on VGG-19 model in order to check the validity of our proposed framework. Both ResNet-50 and VGG-19 model are pre-trained on the ImageNet dataset. We publicly shared our proposed augmented dataset on Kaggle website (https://www.kaggle.com/tr1gg3rtrash/balanced-augmented-covid-cxr-dataset), so that any research community can widely utilize this dataset. Our code is available on GitHub website online (https://github.com/MrinalTyagi/SVD-CLAHE-and-BWCCE).

High-temperature augmented neighborhood metric learning for cross-domain fault diagnosis with imbalanced data

The class imbalance in the machine monitoring data and the data distribution discrepancy due to the different working conditions make it difficult to build a strong mechanical fault diagnosis model in practice. In this paper, we proposed a high-temperature augmented neighborhood metric-learning network (HNMN) for cross-domain fault diagnosis with imbalanced data. In the proposed network, fault types are classified by comparing the distance from the sample to the prototype. And neighborhood component analysis-based domain adaptation scheme was presented to reduce the distance of similar samples in source and target domains thus eliminating the domain shift and solving the variable working condition problem. A classification boundary optimization strategy based on high-temperature mechanism was applied to avoid the overlap of feature distributions of minor classes and major classes, and finally solve the class imbalance problem. The performance and superiority of HNMN are evaluated and analyzed using three fault diagnosis cases. The experimental results demonstrate that the proposed network can achieve higher fault diagnosis accuracy compared to the state-of-the-art methods in the class imbalance and variable working condition diagnosis scenarios.

IMLBoost for intelligent diagnosis with imbalanced medical records

Class imbalance of medical records is a critical challenge for disease classification in intelligent diagnosis. Existing machine learning algorithms usually assign equal weights to all classes, which may reduce classification accuracy of imbalanced records. In this paper, a new Imbalance Lessened Boosting (IMLBoost) algorithm is proposed to better classify imbalanced medical records, highlighting the contribution of samples in minor classes as well as hard and boundary samples. A tailored Cost-Fitting Loss (CFL) function is proposed to assign befitting costs to these critical samples. The first and second derivations of the CFL are then derived and embedded into the classical XGBoost framework. In addition, some feature analysis skills are utilized to further improve performance of the IMLBoost, which also can speed up the model training. Experimental results on five UCI imbalanced medical datasets have demonstrated the effectiveness of the proposed algorithm. Compared with other existing classification methods, IMLBoost has improved the classification performance in terms of F1-score, G-mean and AUC.

DuETNet: Dual Encoder based Transfer Network for thoracic disease classification

• Practical way of achieving memory efficiency under restricted resources is proposed. • Resolution calibration guideline with regard to input image sizes is given. • Entropy based label smoothing considering high class imbalance is proposed. • Channel- and spatial-wise attention modules use query, key, and value matrices. • The attention modules are highly compatible with CNN architectures. In thoracic disease classification , the original chest X-ray images are high resolution images. Nevertheless, in existing convolution neural network (CNN) models, the original images are resized to 224 × 224 before use. Diseases in local areas may not be sufficiently represented because the chest X-ray images have been resized, which compresses information excessively. Therefore, a higher resolution is required to focus on the local representations. Although the large input resolution reduces memory efficiency, previous studies have investigated using CNNs with the large input for classification performance improvement. Moreover, optimization for imbalanced classes is required because chest X-ray images have highly imbalanced pathology labels. Hence, this study proposes the Dual Encoder based Transfer Network (DuETNet) to counter the inefficiency caused by large input resolution and improve classification performance by adjusting the input size based on the RandomResizedCrop method. This image transformation method crops a random area of a given image and resizes it to a given size. Thus, a resolution calibration guideline is a practical way to achieve memory efficiency and performance gains under restricted resources by adjusting the scale factor σ on the training and test images. To treat high class imbalance, we propose entropy based label smoothing method. The method enhances generalization performance for the imbalanced minor classes by penalizing the major classes. The dual encoder comprises channel and spatial encoders, which apply channel- and spatial-wise attention to enhance the relatively significant features from the adjusted images. To evaluate the performance of DuETNet, we used the ChestX-ray14 and MIMIC-CXR-JPG datasets, and DuETNet achieved a new state-of-the-art method.

객체 수준 데이터 증강을 활용한 딥러닝 기반의 강판 표면 결함 검출

The detection of surface defects in the steel manufacturing process is a very important issue in terms of quality assurance and facility management. Recently, many studies have been conducted to improve the performance of detection by applying deep learning models. However, it is still difficult to detect surface defects because of class imbalance problem. In this study, we propose a deep learning model using object-level data augmentation method to improve the detection accuracy of steel sheet surface defect images with class imbalance. We demonstrated the usefulness and applicability of the proposed method by conducting various experiments using Severstal steel defect detection dataset. We applied various deep learning model architectures to the proposed method and proved that its performance was improved compared to the model using existing image-level data augmentation method. Further, the experimental results showed that the proposed method especially improves the defect detection performance of minor classes.

Event recognition method based on dual-augmentation for a Φ-OTDR system with a few training samples.

Thanks to the development of machine learning and deep learning, data-driven pattern recognition based on neural network is a trend for Φ-OTDR system intrusion event recognition. The data-driven pattern recognition needs a large number of samples for training. However, in some scenarios, intrusion signals are difficult to collect, resulting in the lack of training samples. At the same time, labeling a large number of samples is also a very time-consuming work. This paper presents a few-shot learning classification method based on time series transfer and cycle generative adversarial network (CycleGAN) data augmentation for Φ-OTDR system. By expanding the rare samples based on time series transfer and CycleGAN, the number of samples in the dataset can finally meet the requirement of network training. The experimental result shows that even when the training set has two minor classes with only two samples, the average accuracy of the validation set with 5 classification tasks can still reach 90.84%, and the classification accuracy of minor classes can reach 79.28% with the proposed method.

Cross-domain intelligent bearing fault diagnosis under class imbalanced samples via transfer residual network augmented with explicit weight self-assignment strategy based on meta data

Intelligent fault diagnosis methods are significant to mitigate the dependency on expert knowledge and the cost. For the limited faulty data and variational working conditions of real scenarios, cross-domain diagnosis using existing diagnosis models is widely discussed. Especially, methods based on cross-domain transfer learning show great potentiality. However, the class imbalanced data of actual working conditions make it difficult to learn the actual fault feature distribution. To this end, a transfer residual network augmented with explicit weight self-assignment strategy based on meta data(TRN-EWM) is proposed. Specifically, we use a domain-shared ResNet to extract depth features of the data, which effectively avoid gradient disappearance and improve classification performance. Then, to lessen diagnosis difficulties in cross-domain and fully mine the actual feature distribution of the samples, a class imbalanced cross-domain transfer method is carried out. Ultimately, we creatively construct an explicit weight self-assignment strategy based on meta data for sample weight rebalancing, which prevents the dominance of major classes and the overfitting of minor classes. Two transfer experiments are conducted, and average cross-domain diagnosis accuracy of 99.60% is achieved by the proposed method, which shows the effectiveness in bearing fault diagnosis.

Device Free Localization Using Prior Correction-Based Logistic Regression

In this letter, we focus on a device free localization (DFL) system which is based on logistic regression model using received signal strength (RSS) measurements in each link as training and testing data. To reduce the computational complexity of the logistic regression model prediction and improve the localization accuracy by reduction of the redundant training features, we obtain principal component using the measured RSS data through principal component analysis (PCA). Moreover, to overcome the imbalanced data problem, the number of training data belonging to major class is decreased through the sampling technique based on the distance between the reference points. The modified logistic regression model is derived from Bayesian formula based prior correction method due to the change in the ratio of the major and the minor classes in sampled training data. By substituting the transformed principal components of testing dataset to the probability model with the optimal regression coefficients, the localization probabilities of a target are computed at every reference point.

Phytochemical Constituents and Antiproliferative Activities of Essential Oils from Four Varieties of Malaysian Zingiber officinale Roscoe against Human Cervical Cancer Cell Line.

This study evaluates the volatile metabolic constituents and anticancer potential of essential oils distilled from the rhizomes of four Malaysian Zingiber officinale Roscoe (Zingiberaceae family) varieties (Bentong (BE), Cameron Highlands (CH), Sabah (SA), and Bara (BA)). The ginger essential oils were analyzed by gas chromatography coupled with quadrupole mass spectrometry (GC qMS). A total of 58 secondary compounds were tentatively identified, representing 82.6–87.4% of the total ion count. These metabolites comprise mainly of monoterpene hydrocarbons (19.7–25.5%), oxygenated monoterpenes (23.6–33.7%), sesquiterpene hydrocarbons (21.3–35.6%), oxygenated sesquiterpenes (1.5–3.9%), and other minor classes of compounds (0.7–2.7%). Principal component analysis (PCA) enabled differentiation of the analyzed ginger essential oils according to their varieties, with respect to their metabolites and relative quantities. The antiproliferative activity against the HeLa cervical cancer cell line was investigated via the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay. The oils were found to exhibit strong antiproliferative activities with IC50 values of 23.8, 35.3, 41.3, and 42.5 μg/mL for BA, BE, SA, and CH, respectively. These findings suggest that the differences among the secondary metabolites and their abundance in different varieties of Z. officinale essential oils appear to be related to their antiproliferative potential. The strong antiproliferative effects of these oils signified their potential in the prevention and chemotherapy of cervical carcinoma treatment.

AFD-Net: Apple Foliar Disease multi classification using deep learning on plant pathology dataset

BackgroundPlant diseases significantly affect the crop, so their identification is very important. Correct identification of these diseases is crucial for establishing a good disease control strategy to avoid time and financial losses. In general, machines can greatly reduce the possibility of human error. In particular, computer vision techniques developed through deep learning have paved a way to detect and diagnose these plant diseases on the leaf.MethodsIn this work, the model AFD-Net was developed to detect and identify various leaf diseases in apple trees. The dataset is from Kaggle 2020 and 2021 and was financially supported by the Cornell Initiative for Digital Agriculture. An AFD-Net was proposed for leaf disease classification in apple trees and the results of the efficiency of the model are compared with other state-of-the-art deep learning approaches.ResultsThe results of the experiments in the validation dataset show that the proposed AFD-Net model achieves the highest values of 98.7% accuracy for Plant Pathology 2020 and 92.6% for Plant Pathology 2021 compared to other deep learning models in the original and extended datasets.DiscussionThe results also indicate the efficiency of the proposed model in identifying leaf diseases on apple trees for major and minor classes, i.e., for multiple classification.

Novel use of a virtual driving assessment to classify driver skill at the time of licensure

Motor vehicle crash rates are highest immediately after licensure, and driver error is one of the leading causes. Yet, few studies have quantified driving skills at the time of licensure, making it difficult to identify at-risk drivers before independent driving. Using data from a virtual driving assessment implemented into the licensing workflow in Ohio, this study presents the first population-level study classifying degree of skill at the time of licensure and validating these against a measure of on-road performance: license exam outcomes. Principal component and cluster analysis of 33,249 virtual driving assessments identified 20 Skill Clusters that were then grouped into 4 major summary “Driving Classes”; i) No Issues (i.e. careful and skilled drivers); ii) Minor Issues (i.e. an average new driver with minor vehicle control skill deficits); iii) Major Issues (i.e. drivers with more control issues and who take more risks); and iv) Major Issues with Aggression (i.e. drivers with even more control issues and more reckless and risk-taking behavior). Category labels were determined based on patterns of VDA skill deficits alone (i.e. agnostic of the license examination outcome). These Skill Clusters and Driving Classes had different distributions by sex and age, reflecting age-related licensing policies (i.e. those under 18 and subject to GDL and driver education and training), and were differentially associated with subsequent performance on the on-road licensing examination (showing criterion validity). The No Issues and Minor Issues classes had lower than average odds of failing, and the other two more problematic Driving Classes had higher odds of failing. Thus, this study showed that license applicants can be classified based on their driving skills at the time of licensure. Future studies will validate these Skill Cluster classes in relation to their prediction of post-licensure crash outcomes.

Image classification of root-trimmed garlic using multi-label and multi-class classification with deep convolutional neural network

Garlic root trimming is one of the most tedious tasks in the postharvest processing of garlic since the evaluation of trimming results is done by mainly visual inspection for each label of a garlic bulb. Currently, multi-class classification using a deep convolutional neural network (CNN) can automate the evaluation process. However, it can handle only a single label per a garlic image and cannot be used for evaluation of multi-labels in conventional garlic root trimming practices. This study introduced a modified multi-class model and a multi-label model that utilized CNN to classify two labels of a garlic bulb after root trimming. The first label includes good, bad, untrimmed and scratched classes, and the second label consists of clean and muddy classes. The modified multi-class model achieved a classification accuracy of 82.9% while the multi-label gave a better classification performance of minor classes, with an overall accuracy of 95.2%. With the addition of a background image class, classification accuracies of both multi-class model and multi-label model increased to 91.8% and 98.0%, respectively. The background class significantly enhanced the classification performance of multi-label model when it was deployed to a garlic sorting robot. The utilization of data augmentation, dropout, transfer learning and fine-tuning was confirmed to improve model generalization and performance. Multi-label model is recommended for grading of garlic bulbs with multi-labels of after root trimming. The time to process an image was 0.021 s, which is suitable for a real-time garlic sorting and grading robot. The method shows a potential for development of a smart and fully autonomous robots in the postharvest processing for garlic production.

Cost-Sensitive Broad Learning System for Imbalanced Classification and Its Medical Application

As an effective and efficient discriminative learning method, the broad learning system (BLS) has received increasing attention due to its outstanding performance without large computational resources. The standard BLS is derived under the minimum mean square error (MMSE) criterion, while MMSE is with poor performance when dealing with imbalanced data. However, imbalanced data are widely encountered in real-world applications. To address this issue, a novel cost-sensitive BLS algorithm (CS-BLS) is proposed. In the CS-BLS, many variations can be adopted, and CS-BLS with weighted cross-entropy is analyzed in this paper. Weighted penalty factors are used in CS-BLS to constrain the contribution of each sample in different classes. The samples in minor classes are allocated higher weights to increase their contributions. Four different weight calculation methods are adopted to the CS-BLS, and thus, four CS-BLS methods are proposed: Log-CS-BLS, Lin-CS-BLS, Sqr-CS-BLS, and EN-CS-BLS. Experiments based on artificially imbalanced datasets of MNIST and small NORB are firstly conducted and compared with the standard BLS. The results show that the proposed CS-BLS methods have better generalization and robustness than the standard BLS. Then, experiments on a real ultrasound breast image dataset are conducted, and the results demonstrate that the proposed CS-BLS methods are effective in actual medical diagnosis.

Evaluation of a deep-learning model for multispectral remote sensing of land use and crop classification

High-resolution deep-learning-based remote-sensing imagery analysis has been widely used in land-use and crop-classification mapping. However, the influence of composite feature bands, including complex feature indices arising from different sensors on the backbone, patch size, and predictions in transferable deep models require further testing. The experiments were conducted in six sites in Henan province from 2019 to 2021. This study sought to enable the transfer of classification models across regions and years for Sentinel-2A (10-m resolution) and Gaofen PMS (2-m resolution) imagery. With feature selection and up-sampling of small samples, the performance of UNet++ architecture on five backbones and four patch sizes was examined. Joint loss, mean Intersection over Union (mIoU), and epoch time were analyzed, and the optimal backbone and patch size for both sensors were Timm-RegNetY-320 and 256 × 256, respectively. The overall accuracy and F1 scores of the Sentinel-2A predictions ranged from 96.86% to 97.72% and 71.29% to 80.75%, respectively, compared to 75.34%–97.72% and 54.89%–73.25% for the Gaofen predictions. The accuracies of each site indicated that patch size exerted a greater influence on model performance than the backbone. The feature-selection-based predictions with UNet++ architecture and up-sampling of minor classes demonstrated the capabilities of deep-learning generalization for classifying complex ground objects, offering improved performance compared to the UNet, Deeplab V3+, Random Forest, and Object-Oriented Classification models. In addition to the overall accuracy, confusion matrices, precision, recall, and F1 scores should be evaluated for minor land-cover types. This study contributes to large-scale, dynamic, and near-real-time land-use and crop mapping by integrating deep learning and multi-source remote-sensing imagery

SIFLoc: a self-supervised pre-training method for enhancing the recognition of protein subcellular localization in immunofluorescence microscopic images.

With the rapid growth of high-resolution microscopy imaging data, revealing the subcellular map of human proteins has become a central task in the spatial proteome. The cell atlas of the Human Protein Atlas (HPA) provides precious resources for recognizing subcellular localization patterns at the cell level, and the large-scale annotated data enable learning via advanced deep neural networks. However, the existing predictors still suffer from the imbalanced class distribution and the lack of labeled data for minor classes. Thus, it is necessary to develop new methods for coping with these issues. We leverage the self-supervised learning protocol to address these problems. Especially, we propose a pre-training scheme to enhance the conventional supervised learning framework called SIFLoc. The pre-training is featured by a hybrid data augmentation method and a modified contrastive loss function, aiming to learn good feature representations from microscopic images. The experiments are performed on a large-scale immunofluorescence microscopic image dataset collected from the HPA database. Using the same deep neural networks as the classifier, the model pre-trained via SIFLoc not only outperforms the model without pre-training by a large margin but also shows advantages over the state-of-the-art self-supervised learning methods. Especially, SIFLoc improves the prediction accuracy for minor organelles significantly.