Training Instances Research Articles

An online learning algorithm for a neuro-fuzzy classifier with mixed-attribute data

General fuzzy min–max neural network (GFMMNN) is one of the efficient neuro-fuzzy systems for data classification. However, one of the downsides of its original learning algorithms is the inability to handle and learn directly from mixed-attribute data without using encoding techniques. While categorical feature encoding methods can be used with the GFMMNN learning algorithms, they exhibit many shortcomings. Other improved approaches proposed in the literature are not suitable for online learning algorithms working in the dynamically changing environments without ability to retrain or access full historical data, which are usually required for many real world applications. This paper proposes an extended online learning algorithm for the GFMMNN. The proposed method can handle the datasets with both continuous and categorical features. It uses the change in the entropy values of categorical features of the samples contained in a hyperbox to determine if the current hyperbox can be expanded to include the categorical values of a new training instance. An extended architecture of the original GFMMNN and its new membership function are introduced for mixed-attribute data. Important mathematical properties of the proposed learning algorithms are also presented and proved in this paper. The extensive experiments confirmed superior and stable classification performance of the proposed approach in comparison to other relevant learning algorithms for the GFMM model.

Relational reasoning and generalization using nonsymbolic neural networks.

The notion of equality (identity) is simple and ubiquitous, making it a key case study for broader questions about the representations supporting abstract relational reasoning. Previous work suggested that neural networks were not suitable models of human relational reasoning because they could not represent mathematically identity, the most basic form of equality. We revisit this question. In our experiments, we assess out-of-sample generalization of equality using both arbitrary representations and representations that have been pretrained on separate tasks to imbue them with structure. We find neural networks are able to learn (a) basic equality (mathematical identity), (b) sequential equality problems (learning ABA-patterned sequences) with only positive training instances, and (c) a complex, hierarchical equality problem with only basic equality training instances ("zero-shot" generalization). In the two latter cases, our models perform tasks proposed in previous work to demarcate human-unique symbolic abilities. These results suggest that essential aspects of symbolic reasoning can emerge from data-driven, nonsymbolic learning processes. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Ensemble Classification Model for EV Identification From Smart Meter Recordings

Electric vehicles (EVs) often consume large amounts of energy, and uncoordinated charging of many EVs may lead to grid overload, adversely impacting other customers. Electricity distributors require full visibility on the EV distribution to better manage operation planning of their distribution grid. However, they often have incomplete knowledge of EV presence in their network. Identifying EV customers (charging at home) using smart meter data is a nontrivial task for the grid network and energy scheduling. The difficulties include recognizing charging patterns, balancing the number of EV and non-EV customers during modeling, and building an efficient classification model. In this article, we propose a periodic pattern recognition method to extract useful EV charging patterns. Real world smart meter datasets are unbalanced with few EVs and majority of energy customers are those without EVs. We improve Kmedoids evaluated by dynamic time warping to obtain the representative non-EV training samples so that balanced samples over EV and non-EV customers can be obtained. We develop an ensemble classification model (ECM) by taking advantages of multiple classifiers, in which the optimization consists of obtaining the optimal subset of periodic patterns and the optimal parameters in each classifier and the optimal weights for combining classifiers. The superiority of the proposed ECM is demonstrated in comparison to several baseline models.

Solving large-scale multi-agent tasks via transfer learning with dynamic state representation

Many research results have emerged in the past decade regarding multi-agent reinforcement learning. These include the successful application of asynchronous advantage actor-critic, double deep Q-network and other algorithms in multi-agent environments, and the more representative multi-agent training method based on the classical centralized training distributed execution algorithm QMIX. However, in a large-scale multi-agent environment, training becomes a major challenge due to the exponential growth of the state-action space. In this article, we design a training scheme from small-scale multi-agent training to large-scale multi-agent training. We use the transfer learning method to enable the training of large-scale agents to use the knowledge accumulated by training small-scale agents. We achieve policy transfer between tasks with different numbers of agents by designing a new dynamic state representation network, which uses a self-attention mechanism to capture and represent the local observations of agents. The dynamic state representation network makes it possible to expand the policy model from a few agents (4 agents, 10 agents) task to large-scale agents (16 agents, 50 agents) task. Furthermore, we conducted experiments in the famous real-time strategy game Starcraft II and the multi-agent research platform MAgent. And also set unmanned aerial vehicles trajectory planning simulations. Experimental results show that our approach not only reduces the time consumption of a large number of agent training tasks but also improves the final training performance.

Adaptive Prototypical Networks With Label Words and Joint Representation Learning for Few-Shot Relation Classification.

Relation classification (RC) task is one of fundamental tasks of information extraction, aiming to detect the relation information between entity pairs in unstructured natural language text and generate structured data in the form of entity-relation triple. Although distant supervision methods can effectively alleviate the problem of lack of training data in supervised learning, they also introduce noise into the data and still cannot fundamentally solve the long-tail distribution problem of the training instances. In order to enable the neural network to learn new knowledge through few instances such as humans, this work focuses on few-shot relation classification (FSRC), where a classifier should generalize to new classes that have not been seen in the training set, given only a number of samples for each class. To make full use of the existing information and get a better feature representation for each instance, we propose to encode each class prototype in an adaptive way from two aspects. First, based on the prototypical networks, we propose an adaptive mixture mechanism to add label words to the representation of the class prototype, which, to the best of our knowledge, is the first attempt to integrate the label information into features of the support samples of each class so as to get more interactive class prototypes. Second, to more reasonably measure the distances between samples of each category, we introduce a loss function for joint representation learning (JRL) to encode each support instance in an adaptive manner. Extensive experiments have been conducted on FewRel under different few-shot (FS) settings, and the results show that the proposed adaptive prototypical networks with label words and JRL has not only achieved significant improvements in accuracy but also increased the generalization ability of FSRC.

A probabilistic framework for mutation testing in deep neural networks

Context: Mutation Testing (MT) is an important tool in traditional Software Engineering (SE) white-box testing. It aims to artificially inject faults in a system to evaluate a test suite's capability to detect them, assuming that the test suite defects finding capability will then translate to real faults. If MT has long been used in SE, it is only recently that it started gaining the attention of the Deep Learning (DL) community, with researchers adapting it to improve the testability of DL models and improve the trustworthiness of DL systems. Objective: If several techniques have been proposed for MT, most of them neglected the stochasticity inherent to DL resulting from the training phase. Even the latest MT approaches in DL, which propose to tackle MT through a statistical approach, might give inconsistent results. Indeed, as their statistic is based on a fixed set of sampled training instances, it can lead to different results across instances set when results should be consistent for any instance. Methods: In this work, we propose a Probabilistic Mutation Testing (PMT) approach that alleviates the inconsistency problem and allows for a more consistent decision on whether a mutant is killed or not. Results: We show that PMT effectively allows a more consistent and informed decision on mutations through evaluation using three models and eight mutation operators used in previously proposed MT methods. We also analyze the trade-off between the approximation error and the cost of our method, showing that relatively small error can be achieved for a manageable cost. Conclusion: Our results showed the limitation of current MT practices in DNN and the need to rethink them. We believe PMT is the first step in that direction which effectively removes the lack of consistency across test executions of previous methods caused by the stochasticity of DNN training.

Transfer Learning for Low-Resource Multilingual Relation Classification

Relation classification (sometimes called relation extraction ) requires trustworthy datasets for fine-tuning large language models, as well as for evaluation. Data collection is challenging for Indian languages, because they are syntactically and morphologically diverse, as well as different from resource-rich languages like English. Despite recent interest in deep generative models for Indian languages, relation classification is still not well served by public datasets. In response, we present IndoRE , a dataset with 21K entity- and relation-tagged gold sentences in three Indian languages (Bengali, Hindi, and Telugu), plus English. We start with a multilingual BERT (mBERT)-based system that captures entity span positions and type information, and provides competitive performance on monolingual relation classification. Using this baseline system, we explore transfer mechanisms between languages and the scope to reduce expensive data annotation while achieving reasonable relation extraction performance. Specifically, we (a) study the accuracy-efficiency trade-off between expensive, manually labeled gold instances vs. automatically translated and aligned silver instances to train a relation extractor, (b) device a simple mechanism for budgeted gold data annotation by intelligently converting distant-supervised silver training instances to gold training instances with human annotators using active learning, and finally (c) propose an ensemble model to provide a performance boost over that achieved via limited gold training instances. We release the dataset for future research. 1

Radiogenomic classification for MGMT promoter methylation status using multi-omics fused feature space for least invasive diagnosis through mpMRI scans

Accurate radiogenomic classification of brain tumors is important to improve the standard of diagnosis, prognosis, and treatment planning for patients with glioblastoma. In this study, we propose a novel two-stage MGMT Promoter Methylation Prediction (MGMT-PMP) system that extracts latent features fused with radiomic features predicting the genetic subtype of glioblastoma. A novel fine-tuned deep learning architecture, namely Deep Learning Radiomic Feature Extraction (DLRFE) module, is proposed for latent feature extraction that fuses the quantitative knowledge to the spatial distribution and the size of tumorous structure through radiomic features: (GLCM, HOG, and LBP). The application of the novice rejection algorithm has been found significantly effective in selecting and isolating the negative training instances out of the original dataset. The fused feature vectors are then used for training and testing by k-NN and SVM classifiers. The 2021 RSNA Brain Tumor challenge dataset (BraTS-2021) consists of four structural mpMRIs, viz. fluid-attenuated inversion-recovery, T1-weighted, T1-weighted contrast enhancement, and T2-weighted. We evaluated the classification performance, for the very first time in published form, in terms of measures like accuracy, F1-score, and Matthews correlation coefficient. The Jackknife tenfold cross-validation was used for training and testing BraTS-2021 dataset validation. The highest classification performance is (96.84 ± 0.09)%, (96.08 ± 0.10)%, and (97.44 ± 0.14)% as accuracy, sensitivity, and specificity respectively to detect MGMT methylation status for patients suffering from glioblastoma. Deep learning feature extraction with radiogenomic features, fusing imaging phenotypes and molecular structure, using rejection algorithm has been found to perform outclass capable of detecting MGMT methylation status of glioblastoma patients. The approach relates the genomic variation with radiomic features forming a bridge between two areas of research that may prove useful for clinical treatment planning leading to better outcomes.

Meta-ED: Cross-lingual Event Detection Using Meta-learning for Indian Languages

Lack of annotated data is a major concern in Event Detection (ED) tasks for low-resource languages. Cross-lingual ED seeks to address this issue by transferring information across various languages to improve overall performance. In this article, we propose a method for cross-lingual ED with a few training instances. We present a model agnostic meta-learning approach for few-shot cross-lingual ED that is able to find good parameter initialization and enables fast adaptation to new low-resource languages. We evaluate our model on four Indian languages. The results show that our approach significantly outperforms the base model.

Accuracy Comparison of YOLOv7 and YOLOv4 Regarding Image Annotation Quality for Apple Flower Bud Classification

Object detection is one of the most promising research topics currently, whose application in agriculture, however, can be challenged by the difficulty of annotating complex and crowded scenes. This study presents a brief performance assessment of YOLOv7, the state-of-the-art object detector, in comparison to YOLOv4 for apple flower bud classification using datasets with artificially manipulated image annotation qualities from 100% to 5%. Seven YOLOv7 models were developed and compared to corresponding YOLOv4 models in terms of average precisions (APs) of four apple flower bud growth stages and mean APs (mAPs). Based on the same test dataset, YOLOv7 outperformed YOLOv4 for all growth stages at all training image annotation quality levels. A 0.80 mAP was achieved by YOLOv7 with 100% training image annotation quality, meanwhile a 0.63 mAP was achieved with only 5% training image annotation quality. YOLOv7 improved YOLOv4 APs by 1.52% to 166.48% and mAPs by 3.43% to 53.45%, depending on the apple flower bud growth stage and training image annotation quality. Fewer training instances were required by YOLOv7 than YOLOv4 to achieve the same levels of classification accuracies. The most YOLOv7 AP increase was observed in the training instance number range of roughly 0 to 2000. It was concluded that YOLOv7 is undoubtedly a superior apple flower bud classifier than YOLOv4, especially when training image annotation quality is suboptimal.

APSN: Adversarial Pseudo-Siamese Network for Fake News Stance Detection

Fake news is a longstanding issue that has existed on the social network, whose negative impact has been increasingly recognized since the US presidential election. During the election, numerous fake news about the candidates distributes vastly in the online social networks. Identifying inauthentic news quickly is an essential purpose for this research to enhance the trustworthiness of news in online social networks, which will be the task studied in this paper. The fake news stance detection can contribute to detect a startling amount of fake news, which aims at evaluating the relevance between the headline and text bodies. There exists a significant difference between news article headline and text body, since headlines with several key phrases are usually much shorter than the text bodies. Such an information imbalance challenge may cause serious problems for the stance detection task. Furthermore, news article data in online social networks is usually exposed to various types of noise and can be contaminated, which poses more challenges for the stance detection task. In this paper, we propose a novel fake news stance detection model, namely Adversarial Pseudo-Siamese Network model (APSN), to solve these challenges. With coupled input components with imbalanced parameters, APSN can learn and compute feature vectors and similarity score of news article headlines and text bodies simultaneously. In addition, by adopting adversarial setting, besides the regular training set, a set of noisy training instances will be generated and fed to APSN in the learning process, which can significantly enhance the robustness of the model. Extensive experiments have been conducted on a real-world fake news dataset, and the experimental results reveal that the presented model exceeds compared suspicious information detection models with significant advantages.

Development of machine learning–powered models for prostate cancer HRD prediction.

206 Background: In prostate cancer, homologous recombination deficiency (HRD) is associated with poor prognosis, and sensitivity to DNA damaging agents and DNA damage repair (DDR) inhibitors. As new classes of DDR inhibitors become available, identifying patients with HRD will be critical for treatment selection. Here, we present machine learning (ML)-based models trained to predict HRD status directly from hematoxylin and eosin (H&E) whole slide images (WSI). Methods: ML models were trained to predict and segment cells and tissue regions within the tumor microenvironment (TME) using annotated (N=91,021 annotations) WSI of H&E-stained resections from the cancer genome atlas prostate adenocarcinoma (TCGA PRAD) dataset (N=401) and needle core biopsies from a proprietary dataset (N=1,000). Quantified Human Interpretable Features (HIFs) that describe the TME composition were extracted. Three models were trained to predict HRD status using 373 WSI with known HRD score (TCGA PRAD; train N=259, validation N=76, and test N=38). Two models used input from the TME model: An HIF multivariate logistic regression model, and a graph neural network (GNN) where predictions are based on the complex spatial relationships within the TME. An end-to-end (E2E) multiple instance learning model predicted directly from the WSI. Two cutoffs for HRD were defined using Gaussian Mixture Models, resulting in 99 WSI (train N=72, validation N=18, and test N=9) positive for the Genomic Instability (>16 events) cutoff, and 58 WSI (train N=44, validation N=10, test N=4) positive for the Genomic Instability (>22 events) cutoff. An independent validation set of 45 biopsies and 16 resections from a biobank of metastatic castration resistant prostate cancer with HRD status determined by whole-exome sequencing was compared to ML model H&E-based HRD prediction. Results: In the TCGA test set of resection samples, all three models moderately or strongly predicted HRD status, with the HIF model showing the best performance (AUROC 0.87, sensitivity 0.88, specificity 0.62). The same HIF model performed equally well (AUROC 0.85, Sensitivity 0.93, specificity 0.67) in the resection samples from the independent validation set. However, the model performance went down (AUROC 0.69, sensitivity 0.91, specificity: 0.3) when both resection and needle biopsy samples were included, highlighting the importance of a representative training set to achieve robust performance in a real world setting. Further model training and validation with a more diverse dataset is required to accurately assess the performance of the model on needle biopsies. Conclusions: ML models trained on resection prostate cancer samples performed well in predicting HRD status when applied to the same sample type, demonstrating the potential of ML models to predict genomic biomarkers status in surgical specimens for treatment decision.

Understanding and Applying Deep Learning.

The past 10 years have witnessed an explosion in deep learning neural network model development. The most common perceptual models with vision, speech, and text inputs are not general-purpose AI systems but tools. They automatically extract clues from inputs and compute probabilities of class labels. Successful applications require representative training data, an understanding of the limitations and capabilities of deep learning, and careful attention to a complex development process. The goal of this view is to foster an intuitive understanding of convolutional network deep learning models and how to use them with the goal of engaging a wider creative community. A focus is to make it possible for experts in areas such as health, education, poverty, and agriculture to understand the process of deep learning model development so they can help transition effective solutions to practice.

Digital Technologies in the Agro-Industrial Complex on the Example of the Electronic Platform of Rosselkhozbank svoe-rodnoe.ru

The subject of the study is digital technologies in the sphere of agro-industrial complex on the example of Ecosystem “Svoe” of JSC Rosselkhozbank. The purporse of the work are: analytical assessment of the essence and function of digital technologies in the Russian agricultural industry using economic and statistical methods; identification of the problems that are an obstacle to the introduction of modern technologies; and providing practical recommendations on informing more agro-industrial companies about the resources of the “Svoe’’ ecosystem for their use in their current activities. In recent years, the digitalization processes in the industry have intensified, but not yet reached the required level for more efficient business. In addition, for the most part only large enterprises and agricultural holdings can afford the introduction of innovative products. Big business is ensuring the continuous growth of economic indicators in the industry. Small enterprises digitalization processes due to a lack of financial and human resources are slow. To assist small and medium-sized businesses, the ecosystem “Svoe” of Rosselkhozbank was launched in 2020, which is a high-quality example of the introduction of digitalization in agro-industrial complex. Finally, the conclusions are drawn in relation to the need of state support for digitalization pro­cesses in the agro-industrial sector, both through financial investments in the activities of enterprises and through reforming the educational sphere connected with training, education and retraining of industry representatives using the achievements of scientific and technological progress; participation of the state in the formation of the countryside as an attractive place to live. Some conclusions about the uneven nature of the implementation of digitalization processes at the enterprises of the industry, the need to reform the education system using innovative developments in the training of specialists for agribusiness enterprises, and the inevitability of improving rural areas to attract additional human resources to the industry are drawn. Despite the existence of the “Svoe” ecosystem of Rosselkhozbank for a short time, it is worth noting the huge benefit of the platform in solving many problems of its users.

Instance Selection-Based Surrogate-Assisted Genetic Programming for Feature Learning in Image Classification.

Genetic programming (GP) has been applied to feature learning for image classification and achieved promising results. However, many GP-based feature learning algorithms are computationally expensive due to a large number of expensive fitness evaluations, especially when using a large number of training instances/images. Instance selection aims to select a small subset of training instances, which can reduce the computational cost. Surrogate-assisted evolutionary algorithms often replace expensive fitness evaluations by building surrogate models. This article proposes an instance selection-based surrogate-assisted GP for fast feature learning in image classification. The instance selection method selects multiple small subsets of images from the original training set to form surrogate training sets of different sizes. The proposed approach gradually uses these surrogate training sets to reduce the overall computational cost using a static or dynamic strategy. At each generation, the proposed approach evaluates the entire population on the small surrogate training sets and only evaluates ten current best individuals on the entire training set. The features learned by the proposed approach are fed into linear support vector machines for classification. Extensive experiments show that the proposed approach can not only significantly reduce the computational cost but also improve the generalisation performance over the baseline method, which uses the entire training set for fitness evaluations, on 11 different image datasets. The comparisons with other state-of-the-art GP and non-GP methods further demonstrate the effectiveness of the proposed approach. Further analysis shows that using multiple surrogate training sets in the proposed approach achieves better performance than using a single surrogate training set and using a random instance selection method.

Sensors and machine learning and AI operation-constrained process control method for sensor-aided industrial internet of things and smart factories

The Industrial Internet of Things (IIoT) incorporates intelligent computing and artificial intelligence paradigms in smart factories for ease of production and human-less interventions. The controlling unit is integrated into the smart operations in managing, controlling, and monitoring industrial operations. This article introduces an Operation-Constrained Process Control (OCPC) for preventing time-lag errors between smart machine operation cycles. The proposed method considers the operational cycles, completion time, and output efficiency metrics for identifying time lags and errors in production. Depending on the maximum productivity-based outcomes, further allocations or modifications in the regular operational cycles are identified. The productivity is analyzed based on the previous outcomes using federated learning. This learning implies multi-sensor knowledge update and production efficiency through repeated training. Therefore, the errors in cycle assignment and production completion are synchronized under controlled error. This method is analyzed using time lag, error, production efficiency, and training instances.

Enviromic-based kernels may optimize resource allocation with multi-trait multi-environment genomic prediction for tropical Maize

BackgroundSuccess in any genomic prediction platform is directly dependent on establishing a representative training set. This is a complex task, even in single-trait single-environment conditions and tends to be even more intricated wherein additional information from envirotyping and correlated traits are considered. Here, we aimed to design optimized training sets focused on genomic prediction, considering multi-trait multi-environment trials, and how those methods may increase accuracy reducing phenotyping costs. For that, we considered single-trait multi-environment trials and multi-trait multi-environment trials for three traits: grain yield, plant height, and ear height, two datasets, and two cross-validation schemes. Next, two strategies for designing optimized training sets were conceived, first considering only the genomic by environment by trait interaction (GET), while a second including large-scale environmental data (W, enviromics) as genomic by enviromic by trait interaction (GWT). The effective number of individuals (genotypes × environments × traits) was assumed as those that represent at least 98% of each kernel (GET or GWT) variation, in which those individuals were then selected by a genetic algorithm based on prediction error variance criteria to compose an optimized training set for genomic prediction purposes.ResultsThe combined use of genomic and enviromic data efficiently designs optimized training sets for genomic prediction, improving the response to selection per dollar invested by up to 145% when compared to the model without enviromic data, and even more when compared to cross validation scheme with 70% of training set or pure phenotypic selection. Prediction models that include G × E or enviromic data + G × E yielded better prediction ability.ConclusionsOur findings indicate that a genomic by enviromic by trait interaction kernel associated with genetic algorithms is efficient and can be proposed as a promising approach to designing optimized training sets for genomic prediction when the variance-covariance matrix of traits is available. Additionally, great improvements in the genetic gains per dollar invested were observed, suggesting that a good allocation of resources can be deployed by using the proposed approach.

A novel weakly supervised semantic segmentation framework to improve the resolution of land cover product

Open-source land cover products (LCPs) are essential for many areas of scientific research. However, they have deficiencies such as low accuracy, low resolution, and poor timeliness when applied to a specific area. Therefore, we developed WESUP-LCP, a novel two-stage weakly supervised semantic segmentation framework to improve the resolution of LCPs without any manual labeling cost. In the first stage, we designed a sample transferring module to transfer accurate and representative training samples from low-resolution LCPs. In the second, partially labeled superpixels generated from the transferred labels were used to train the weakly supervised learning network and pseudo-labels were generated through dynamic label propagation during the training process. We designed two experiments to verify the performance of the proposed method, including the experiment on improving the resolution of LCP in specific areas (improving 30 m LCP to 10 m resolution) and the experiment on the public dataset of the multi-temporal semantic change detection task of 2021 IEEE GRSS Data Fusion Contest (improving 30 m LCP to 1 m resolution). The results showed that our method has remarkable advantages over other methods, demonstrating the applicability of WESUP-LCP to improve LCP resolution.

A Multidepth and Multibranch Network for Hyperspectral Target Detection Based on Band Selection

Deep learning (DL) has recently risen to prominence in hyperspectral target detection (HTD). Nevertheless, how to tackle the extreme training sample imbalance together with achieving target highlighting and background suppression is challenging. Additionally, due to the spectral redundancy of hyperspectral imagery (HSI), it is a new course for HTD through band selection (BS) to retain crucial bands thereupon improving the subsequent detection performance. Accordingly, we propose a DL-based BS-HTD (DLBSTD) algorithm, incorporating DL-based BS with DL-based HTD for the first time. Most significantly, a multi-depth and multi-branch network (MDBN) for HTD based on a novel BS method is proposed. First of all, the BS method including an alternating local-global reconstruction network (ALGRN) and a correlation measurement strategy provides representative bands containing key target information for MDBN. For the training sample imbalance of MDBN, we develop a BS-based method to select multifarious representative background training samples and propose a target band random substitution (TBRS) strategy to augment an ample target training set. Lastly, the MDBN composed of a multi-depth feature extraction (MDFE) module, three fusion strategies, and the parallel local convolution and gated recurrent unit (Conv-GRU) fully taps the spectral feature relationships to highlight targets and suppress backgrounds. Compared with nine competitive HTD algorithms, we carry out plentiful experiments on four classical datasets exhibiting that the proposed DLBSTD has strong generalization and salient detection performance of target highlighting and background suppression.

Hyperspectral Target Detection via Spectral Aggregation and Separation Network With Target Band Random Mask

Hyperspectral target detection (HTD) is a pixel-wise detection method based on limited prior targets and spectral differences, which has been widely studied and applied in many fields. Recently, deep learning (DL) plays an important role in hyperspectral imagery (HSI) processing. However, for HTD, the severe lack of class-balanced training sets is an enormous challenge. Meanwhile, it is difficult to suppress backgrounds while highlighting targets through the deep network. To address these issues, we propose a spectral aggregation and separation network (SASN) with a target band random mask (TBRM) for HTD in this paper. For the training sets of SASN, a multifarious representative background selection strategy (MRBS) is first proposed to obtain a multifarious and representative background training set. Next, aiming at the notorious class imbalance, a data augmentation (DA) method, TBRM, is proposed to generate adequate target training set by repeating randomly zero-masking the spectral bands of a prior target. Subsequently, in the training of SASN, residual connection and squeeze-and-excitation (SE) channel attention mechanism are applied to fully extract high discriminative features and nonlinear ones in the spectra. Besides, to better separate the targets and backgrounds, a triplet-soft loss function is presented, which makes the training in the direction of spectral separation of background samples from both the prior target and target samples. During testing, the trained SASN distinguishes the spectral similarities and differences simultaneously for highlighting targets and suppressing backgrounds. Moreover, extensive experimental results validate that the proposed method has superior detection performances, background suppression capacity, and separability compared with ten cutting-edge HTD algorithms on six benchmark HSI datasets.