Training Sample Selection Method Research Articles

CLFR-Det: Cross-level feature refinement detector for tiny-ship detection in SAR images

CLFR-Det: Cross-level feature refinement detector for tiny-ship detection in SAR images

Hourly Building Energy Consumption Prediction Using a Training Sample Selection Method Based on Key Feature Search

For the management of building operations, hourly building energy consumption prediction (HBECP) is critical. Many factors, such as energy types, expected day intervals, and acquired feature types, significantly impact HBECP. However, the existing training sample selection methods, especially during transitional seasons, are unable to properly adapt to changes in operational conditions. The key feature search selection (KFSS) approach is proposed in this study. This technique ensures a quick response to changes in the parameters of the predicted day while enhancing the model’s accuracy, stability, and generalization. The best training sample set is found dynamically based on the similarity between the feature on the projected day and the historical data, and feature scenario analysis is used to make the most of the acquired data features. The hourly actual data in two years are applied to a major office building in Zhuhai, China as a case study. The findings reveal that, as compared to the original methods, the KFSS method can track daily load well and considerably enhance prediction accuracy. The suggested training sample selection approach can enhance the accuracy of prediction days by 14.5% in spring and 4.9% in autumn, according to the results. The proposed feature search and feature extraction strategy are valuable for enhancing the robustness of data-driven models for HBECP.

An active learning Gaussian modeling based multi-objective evolutionary algorithm using population guided weight vector evolution strategy.

The inverse model based multi-objective evolutionary algorithm (IM-MOEA) generates offspring by establishing probabilistic models and sampling by the model, which is a new computing schema to replace crossover in MOEAs. In this paper, an active learning Gaussian modeling based multi-objective evolutionary algorithm using population guided weight vector evolution strategy (ALGM-MOEA) is proposed. To properly cope with multi-objective problems with different shapes of Pareto front (PF), a novel population guided weight vector evolution strategy is proposed to dynamically adjust search directions according to the distribution of generated PF. Moreover, in order to enhance the search efficiency and prediction accuracy, an active learning based training sample selection method is designed to build Gaussian process based inverse models, which chooses individuals with the maximum amount of information to effectively enhance the prediction accuracy of the inverse model. The experimental results demonstrate the competitiveness of the proposed ALGM-MOEA on benchmark problems with various shapes of Pareto front.

A training sample selection method for predicting software defects

A training sample selection method for predicting software defects

A fast and efficient green apple object detection model based on Foveabox

Fruit object detection is crucial for automatic harvesting systems, serving applications such as orchard yield measurement and fruit harvesting. In order to achieve fast recognition and localization of green apples and meet the real-time working requirements of the vision system of harvesting robots, a fast optimized Foveabox detection model (Fast-FDM) is proposed. Fast-FDM uses an optimized form of anchor-free Foveabox to accurately and efficiently detect green apples in harvesting environments. Specifically, the EfficientNetV2-S with fast training and small size is used as the backbone network, a weighted bi-directional feature pyramid network (BiFPN) is employed as the feature extraction network to fuse multi-scale features easily and fast, and then the fused features are fed to the fovea head prediction network for the classification and bounding box prediction. Furthermore, an adaptive training sample selection (ATSS) method is adopted to directly select positive and negative samples, allowing green fruits of different scales to obtain higher recall and achieve more accurate green apple detection. Experimental results show that the proposed Fast-FDM realizes a mean average precision (mAP) of 62.3% for green apple detection using fewer parameters and floating point of operations (FLOPs), achieving better trade-offs between accuracy and detection efficiency.

A Novel Knowledge-Aided Training Samples Selection Method for Terrain Clutter Suppression in Hybrid Baseline Radar Systems

For a space-based radar system with hybrid baseline, the problem of clutter angle-Doppler spectral broadening poses a significant challenge to clutter cancellation in the terrain fluctuant observation scene. To improve the robustness of clutter suppression, this paper proposed a homogeneous sample selection method based on a novel concept of Generalized Spatial Spectrum Density Function (GSSDF). This method can be summarized as three crucial steps. First, the GSSDF was constructed by the prior information of DEM data, radar system parameters, and the backscattering model. Then, the angle of deflection (AOD) and the equivalent bandwidth (EBW) of GSSDF were adopted to measure the diffusion of clutter spectral. Subsequently, the sample selection criterion was established by a new threshold detection strategy. But the key here is that an appropriate detection threshold of the AOD and EBW can be determined by the characteristic of filter response. To summarize, this approach ensures that the training samples sharing similar clutter properties can be selected to estimate clutter covariance matrix (CCM); thereby, enhances clutter suppression capability. Finally, the experimental results demonstrated that the proposed method can obtain better clutter suppression performance than other contrast methods.

A training sample selection method based on united generalised inner product statistics for STAP

A training sample selection method based on united generalised inner product statistics for STAP

Covariance Matrix Whitening-Based Training Sample Selection Method for Airborne Radar

As training samples are not always target-free in space-time processing for airborne radar, the traditional methods usually use the sample covariance matrix (SCM) as the test covariance matrix (TCM) to censor contaminated training samples. However, the SCM cannot represent the property of the cell under test (CUT) accurately, resulting in low selection efficiency. To deal with this problem, this letter proposes a novel training sample selection method based on covariance matrix whitening. Specifically, we utilize the reconstructed subaperture's clutter covariance matrix (RSCCM) of the CUT as the TCM. The RSCCM is only determined by the CUT and can characterize the CUT directly. Then, we use the RSCCM to whiten the subaperture's covariance matrix of the training sample. A criterion for selecting the training samples is derived based on the maximum eigenvalue of the whitened subaperture's covariance matrix, which is related to the energy of the outliers and more stable than the statistic of the generalized inner product method. Simulations are conducted to evaluate the performance of the proposed method.

Information Geometric Means-Based STAP for Nonhomogeneous Clutter Suppression in High Frequency Hybrid Sky-Surface Wave Radar

The nonhomogeneous clutter is one of the biggest challenge for ocean remote sensing radar system such as high frequency hybrid sky-surface wave radar. Space-time adaptive processing (STAP) is a powerful method to suppress the clutter but the training samples in adjacent range gates are not always independent and identically distributed with the cell under test in the heterogeneous environment. So the estimation of the clutter covariance matrix will be inaccurate and degrade the STAP performance. In this paper, a novel training sample selection method based on information geometric means is proposed to choose the more homogeneous secondary data containing the precise clutter information of cell under test. The simulation results of selection performance is conducted to show the advantages of the Information geometric mean based estimators compared with the existing estimators. Combined with joint domain localized STAP, the effectiveness of the proposed method is verified by experimental data. The results show the proposed method outperforms conventional STAP method and the nonhomogeneous clutter can be greatly suppressed.

Poststack Seismic Data Denoising Based on 3-D Convolutional Neural Network

Deep learning has been successfully applied to image denoising. In this study, we take one step forward by using deep learning to suppress random noise in poststack seismic data from the aspects of network architecture and training samples. On the one hand, poststack seismic data denoising mainly aims at 3-D seismic data. We designed an end-to-end 3-D denoising convolutional neural network (3-D-DnCNN) that takes raw 3-D cubes as input in order to better extract the features of the 3-D spatial structure of poststack seismic data. On the other hand, denoising images with deep learning require noisy–clean sample pairs for training. In the field of seismic data processing, researchers usually try their best to suppress noise by using complex processes that combine different methods, but clean labels of seismic data are not available. In addition, building training samples in field seismic data has become an interesting but challenging problem. Therefore, we propose a training sample selection method that contains a complex workflow to produce comparatively ideal training samples. Experiments in this study demonstrate that deep learning can directly learn the ability to denoise field seismic data from selected samples. Although the building of the training samples may occur through a complex process, the experimental results of synthetic seismic data and field seismic data show that the 3-D-DnCNN has learned the ability to suppress the Gaussian noise and super-Gaussian noise from different training samples. Moreover, the 3-D-DnCNN network has better denoising performance toward arc-like imaging noise. In addition, we adopt residual learning and batch normalization in order to accelerate the training speed. After network training is satisfactorily completed, its processing efficiency can be significantly higher than that of conventional denoising methods.

A Novel Sample Selection Method for Impervious Surface Area Mapping Using JL1-3B Nighttime Light and Sentinel-2 Imagery

Urbanization has attracted wide and active interests due to the impact on regional sustainable development. As an important indicator of urbanization, impervious surface area (ISA) should be accurately monitored. In scenario of identifying ISA by supervised classification from satellite images, the training samples are usually labeled manually, which is highly labor-intensive and time-consuming. High-resolution nighttime light image provides a unique footprint of human activities and settlements which are strongly correlated with ISA. In view of this, a novel ISA training sample selection method is proposed by integrating the JL1-3B high-resolution nighttime light imagery and Sentinel-2 time series imagery, and the random forest is applied to classify ISA from Sentinel-2 imagery. The quality of the automatically selected samples was quantitatively validated. There were over three study areas, and the overall classification accuracies were above 97%, showing reliable and robust performance. Compared with conventional methods, the proposed approach achieves satisfactory results in separating bare land from ISA. This study provides a data fusion way which can automatically generate sufficient and high-quality training samples for ISA mapping, and suggests that high-resolution nighttime imagery could demonstrate a promising potential for urban remote sensing.

Heterogeneous Transfer Learning for Hyperspectral Image Classification Based on Convolutional Neural Network

Deep convolutional neural networks (CNNs) have shown their outstanding performance in the hyperspectral image (HSI) classification. The success of CNN-based HSI classification relies on the availability sufficient training samples. However, the collection of training samples is expensive and time consuming. Besides, there are many pretrained models on large-scale data sets, which extract the general and discriminative features. The proper reusage of low-level and midlevel representations will significantly improve the HSI classification accuracy. The large-scale ImageNet data set has three channels, but HSI contains hundreds of channels. Therefore, there are several difficulties to simply adapt the pretrained models for the classification of HSIs. In this article, heterogeneous transfer learning for HSI classification is proposed. First, a mapping layer is used to handle the issue of having different numbers of channels. Then, the model architectures and weights of the CNN trained on the ImageNet data sets are used to initialize the model and weights of the HSI classification network. Finally, a well-designed neural network is used to perform the HSI classification task. Furthermore, attention mechanism is used to adjust the feature maps due to the difference between the heterogeneous data sets. Moreover, controlled random sampling is used as another training sample selection method to test the effectiveness of the proposed methods. Experimental results on four popular hyperspectral data sets with two training sample selection strategies show that the transferred CNN obtains better classification accuracy than that of state-of-the-art methods. In addition, the idea of heterogeneous transfer learning may open a new window for further research.

A graded proportion method of training sample selection for updating conventional soil maps

Selection of training samples is a vital step in updating conventional soil maps when utilizing data mining models. Quality of training samples significantly affects the mapping results and accuracies of the updated soil maps. The area-weighted proportion method was a common method for generating training samples. However, this method usually assigns too small weight to those soil types of small areas and large weight to those of large areas in sample size allocation, which causes the unreasonable proportions of sample numbers for soil types and thereby biases the representation of soil-environmental relationships for those soil types. Meanwhile, random selection of training samples from a soil type may generate some ‘noise’ samples located in the transition areas between soil types. These two aspects in training sample selection could probably reduce the accuracy of the updated soil maps. In this study, a new method was developed to select training samples based on soil type grading according to their area coverages. The method consists of two steps. The first step is to determine the numbers of training samples for each soil type based on soil type grading so as to maintain the reasonable proportion in sample numbers among soil types with different area coverages. The second step is to select typical (representative) samples for each soil type from conventional soil map, to avoid generation of ‘noise samples’. To evaluate the proposed method, the method was compared with three other training sample selection methods with four training sample sizes. Each method was ran for 100 times to generate training sample datasets with each sample size to evaluate their effectiveness and stability. Random forest was employed to generate updated soil maps in a small watershed in Raffelson, Wisconsin (USA). The validation results showed that the graded proportion method effectively solved the imbalanced issue of training samples among soil types with area coverages in big differences caused by the area-weighted proportion strategy. Thus training samples generated using the proposed method usually obtained more accurate and reasonable mapping results than those using the area-weighted proportion strategy. Furthermore, the performance of the proposed method was more stable than that of the area-weighted proportion strategy with the training sample size increasing. It is concluded that the proposed method is an effective training sample selection method for data mining model to update conventional soil maps.

Sample selection-based hierarchical extreme learning machine

Abstract Large amounts of training data in machine learning can keep the accuracy high to a certain extent, but the time costs are high because of the exorbitant amount of data and their dimensionality. Therefore, how to simultaneously select the most useful training data set and extract the main features of the samples, especially for image data, are essential problems that urgently need to be solved in the field of large-scale machine learning. Herein, a training sample selection method that is based on the fuzzy c-means clustering algorithm (FCM) is proposed for the problems. It first utilises condensed nearest neighbour (CNN) to make a preliminary selection of training samples. Then, it utilises the FCM to get the centres of the selected data, and, finally, it effectively condenses the sample using a compression parameter. Meanwhile, considering the critical influence of the sample features on the classification model, this paper selects the hierarchical extreme learning machine (H-ELM) model to better solve the classification task. Based on this, the paper presents the FCM-CNN-H-ELM framework for data classification, which combines FCM-Based CNN and H-ELM. The results of the experiments show that the proposed training sample selection method and classification framework can guarantee consistent, even higher, prediction results with a small number of training samples, and significantly reduce the training time.

Training sample selection for space–time adaptive processing based on multi‐frames

As training samples are not identically distributed with cell under test (CUT) in heterogeneous environments, the performance of space–time adaptive processing (STAP) to suppress clutter degrades. To improve the performance of STAP, this study proposes a novel training sample selection method for STAP based on multi-frames. The key feature of the new method is to overcome the deficit of independent and identically distributed (IID) training samples in heterogeneous environments. First, multi-frames are selected according to similarity from continuous frames; second, training samples which are IID to the CUT from current frame are selected; finally, combine the training samples of the current frame and same range samples in reference frame into final training sample set. The proposed method is applied to real-radar data, and experimental results demonstrate the effectiveness of the proposed method.

A training samples selection method based on system identification for STAP

Abstract In space-time adaptive processing (STAP), the selected training samples should have the same covariance matrix as the clutter of the cell under test (CUT). The traditional methods usually select samples whose waveforms are similar to that of the CUT. We notice that completely dissimilar waveforms may have the same covariance matrix. As a result, many valid samples are lost in traditional methods. So we propose a training samples selection method based on system identification. The proposed methods select samples with similar covariance matrices instead of similar waveforms. First, a samples selection model based on system identification is proposed. Then, the neural network is used to identify the clutter model of the CUT. Finally, samples are selected according to the output variance. Compared with the methods in [1, 2, 3, 4], the proposed method has the following advantages: (1) More than twice the valid training samples can be obtained; (2) The clutter suppression performance can be improved more than 2 dB for the measured data.

Training samples selection method for space‐time adaptive processing based on clutter covariance matrix reconstruction

A novel training samples selection method is proposed to improve the performance of space-time adaptive processing in non-homogeneous environments, where training samples are contaminated by interference target signals (outliers). First, the outlier component is removed from the sample covariance matrix based on clutter covariance matrix reconstruction, which utilises the clutter Capon spectrum integrated over a sector separated from the location of outlier. Secondly, the integral is calculated approximately by discrete sum method. Finally, the reconstructed clutter covariance matrix is combined with the generalised inner products algorithm to form the proposed statistics and the contaminated training samples are eliminated. Simulation results validate the effectiveness of the proposed method.

A method for selecting training samples based on camera response

In the process of spectral reflectance reconstruction, sample selection plays an important role in the accuracy of the constructed model and in reconstruction effects. In this paper, a method for training sample selection based on camera response is proposed. It has been proved that the camera response value has a close correlation with the spectral reflectance. Consequently, in this paper we adopt the technique of drawing a sphere in camera response value space to select the training samples which have a higher correlation with the test samples. In addition, the Wiener estimation method is used to reconstruct the spectral reflectance. Finally, we find that the method of sample selection based on camera response value has the smallest color difference and root mean square error after reconstruction compared to the method using the full set of Munsell color charts, the Mohammadi training sample selection method, and the stratified sampling method. Moreover, the goodness of fit coefficient of this method is also the highest among the four sample selection methods. Taking all the factors mentioned above into consideration, the method of training sample selection based on camera response value enhances the reconstruction accuracy from both the colorimetric and spectral perspectives.

The Training Method of General Regression Neural Network for GDOP Approximation

Aimed at the problem of real-time and accurate Geometry Dilution of Precision (GDOP) approximation, a new method using general regression neural network (GRNN) was proposed firstly, and the training samples selection and normalization method was studied by using spectrum analysis. The computation results show that the symmetrical constellation needs 24 hours continuous samples while the hybrid one needs 72 hours to train the neural network sufficiently. Finally, the performance analysis shows that this new method has excellent performance on temporal and spatial generalization approximation accuracy, when trained GRNN are used, the GDOP computational error is less than 0.25 within 30 days, and the error is less than 0.3 within 10 degrees latitude/longitude area.