Deep reinforcement learning-based patch selection for illuminant estimation

Using convolutional neural network models illumination estimation according to light colors

This paper presents a hybrid approach for estimating the admittance matrix of open-loop distribution networks by integrating Total Least Squares (TLS) regression with deep learning (DL) and graph neural networks (GNNs). The proposed model refines initial TLS estimates by leveraging deep learning to handle nonlinearities and graph neural networks to incorporate network topology, thereby addressing the challenges posed by noisy and incomplete data. Physical constraints-including Kirchhoff's current and voltage laws-are embedded within the optimization process to ensure that the estimated admittance matrix remains physically consistent. Extensive testing on the IEEE 33-bus and 123-bus systems demonstrates significant improvements in estimation accuracy, convergence speed, and adherence to physical principles, underscoring the model's potential for real-world applications in modern power system analysis and operation.

Automated invasive ductal carcinoma detection based using deep transfer learning with whole-slide images

The Zeta-image, illuminant estimation, and specularity manipulation

Large scale near-duplicate image retrieval (NDIR) relies on the Bag-of-Words methodology which quantizes local features into visual words. However, the direct match of these visual words typically leads to unpleasant mismatches due to quantization errors. To enhance the discriminability of the matching process, existing methods usually exploit hand-crafted contextual information, which has limited performance in complicated real-world scenarios. In contrast, we in this paper propose a trainable lightweight embedding network to extract local binary features. The network takes image patches as inputs and generates the binary code that can be efficiently stored in the inverted indexing file and helps discard mismatches immediately during the retrieval process. We improve the discriminability of the code by elaborately composing the training patches for network optimization, which consists of a proper inter-class (non-duplicate) patches selection and a rich intra-class (near-duplicate) patch generation. We evaluate our approach on the open NDIR dataset, INRIA CopyDays, and the experimental results show that our method performs favorably against the state-of-the-art algorithms. Furthermore, with a relatively short code length, our approach achieves higher query speed and lower storage occupation.

<p>In recent years, disasters are more frequent and enormous due to global warming. In the field of hydrology, high-precision rainfall-runoff modeling is required. Recently, deep learning has been applied to rainfall-runoff modeling and shows high accuracy. It is also expected that the accuracy will be improved by using ensemble learning for deep learning. This study tried to improve the accuracy of river flow estimation by performing ensemble learning for deep learning. Stacking was used as the ensemble learning method. For deep learning, LSTM, CNN, and MLP was used and compared. XGBoost was used as the learning device used for ensemble learning. The target area was the Tedori River basin in Ishikawa Prefecture, Japan. In deep learning, the input data were daily average precipitation and temperature. In deep learning and ensemble learning, the target data was the daily average river flow. RMSE was used as the evaluation index. As a result, the accuracy was the highest after ensemble learning when using LSTM. It shows that the selection of the learning device is important for ensemble learning.</p>

Illumination estimation is a fundamental prerequisite for many computer vision applications. Various statistics and deep learning-based estimation methods have been proposed, and further studies are ongoing. In this study, we first perform a comparative analysis of representative statistics and deep learning-based methods and subsequently investigate combining them to improve the illumination estimation accuracy. We use hyperspectral images as the training data and support vector regression to combine the methods. Based on the results, we confirm that their combination enhances their accuracy.

Improving battery state estimation accuracy through the addition of a series capacitor

Infrared and visible image fusion (IVIF) is an essential branch of image fusion, and enhancing the visible image of IVIF can significantly improve the fusion performance. However, many existing low‐light enhancement methods are unsuitable for the visible image enhancement of IVIF. In order to solve this problem, this paper proposes a new visible image enhancement method for IVIF. Firstly, the colour balance and contrast enhancement‐based self‐calibrated illumination estimation (CCSCE) is proposed to improve the input image's brightness, contrast, and colour information. Then, the method based on Mutually Guided Image Filtering (muGIF) is adopted to design a strategy to extract details adaptively from the original visible image, which can keep details without introducing additional noise effectively. Finally, the proposed visible image enhancement technique is used for IVIF tasks. In addition, the proposed method can be used for the visible image enhancement of IVIF and other low‐light images. Experiment results on different public datasets and IVIF demonstrate the authors’ method's superiority from both qualitative and quantitative comparisons. The authors’ code will be publicly available at https://github.com/yiqiao666/low‐light‐enhancement‐for‐IVIF/tree/master .

Vein occlusions and diabetic retinopathy are two of many retinal pathologies affecting the retina. Understanding robust vessel segmentation of fundus images is of vital importance for improving the diagnosis results of these diseases. This paper proposes a novel approach for computing the minimum distance for each test patch via the distance comparison within the test patch and cluster centers. The numerous patches are calculated using manual segmentations through the K-means algorithm. We demonstrate the efficiency of learning the simple pattern from each cluster; meanwhile, the mapping function for each cluster is determined by the patches in the training images and their corresponding manual segmentation patches. Two publicly recognized benchmark data sets, namely DRIVE and STARE, are used in our experimental validation. Experimental results show that the proposed approach outperforms conventional methods for vessel segmentation problems validated via public benchmark data sets, i.e., DRIVE and STARE.

State estimation and localization for the autonomous vehicle are essential for accurate navigation and safe maneuvers. The commonly used method is Kalman filtering, but its performance is affected by the noise covariance. An inappropriate set value will decrease the estimation accuracy and even makes the filter diverge. The noise covariance estimation problem has long been considered a tough issue because there is too much uncertainty in where the noise comes from and therefore unable to model it systematically. In recent years, Deep Reinforcement Learning (DRL) has made astonishing progress and is an excellent choice for tackling the problem that cannot be solved by conventional techniques, such as parameter estimation. By finely abstracting the problem as an MDP, we can use the DRL methods to solve it without too many prior assumptions. We propose an adaptive covariance tuning method applied to the Error State Extend Kalman Filter by taking advantage of DRL, called Reinforcement Learning Aided Covariance Tuning. The preliminary experiment result indicates that our method achieves a 14.73% estimation accuracy improvement on average compared with the vanilla fixed-covariance method and bound the estimation error within 0.4 m.

This paper proposes an image patch selection that finds similar patches in multiple images so that image denoising can suppress noise more effectively by exploiting the identified similar patches from the multi-view images. We encode all image patches in multi-view images into a low-dimensional space, and it allows for a denoiser to find similar patches effectively from the space. Our approach enables existing patch-based denoisers, which often find similar patches within an image window, to identify more similar patches by extending the limited search space into the entire space (i.e., all input images). We integrate our technique into state-of-the-art single-view denoising (block-matching and 3D filtering (BM3D)), and demonstrate that the BM3D combined with our approach is able to conduct multi-view image denoising effectively, without a major alteration to the existing algorithm.

S.41-50

Heterogeneous web media consists of many visual concepts, such as objects, scenes and activities, that cannot be semantically decomposed. The task of learning fundamental visual concepts (FVCs) plays an important role in automatically understanding the elements that compose all visual media, as well as in applications of retrieval, annotation, etc. In this paper, we formulate the problem of FVC learning and propose an approach to this problem called neighboring concept distributing (NCD). Our approach models all data using a concept graph, which considers the visual patches in images as nodes and generates the inter-image edges between visual patches in different images and the intra-image edges between visual patches in the same image. The NCD approach distributes semantic information from images to visual patches based on measurements over the concept graph, including fitness, distinctiveness, smoothness and sparseness, without any pre-trained concept detectors or classifiers. We analyze the learnability of the proposed approach and find that, under some conditions, all concepts can be correctly learned with an arbitrarily high probability as the size of the data increases.We demonstrate the performance of the NCD approach using three public datasets. Experimental results show that our approach outperforms state-of-the-art approaches when learning visual concepts from correlated media.

Rice floret number per unit area as one of the key yield structure parameters is directly related to the final yield of rice. Previous studies paid little attention to the effect of the variations in vegetation indices (VIs) caused by rice flowering on rice yield estimation. Unmanned aerial vehicles (UAV) equipped with hyperspectral cameras can provide high spatial and temporal resolution remote sensing data about the rice canopy, providing possibilities for flowering monitoring. In this study, two consecutive years of rice field experiments were conducted to explore the performance of florescence spectral information in improving the accuracy of VIs-based models for yield estimates. First, the florescence ratio reflectance and florescence difference reflectance, as well as their first derivative reflectance, were defined and then their correlations with rice yield were evaluated. It was found that the florescence spectral information at the seventh day of rice flowering showed the highest correlation with the yield. The sensitive bands to yield were centered at 590 nm, 690 nm and 736 nm–748 nm, 760 nm–768 nm for the first derivative florescence difference reflectance, and 704 nm–760 nm for the first derivative florescence ratio reflectance. The florescence ratio index (FRI) and florescence difference index (FDI) were developed and their abilities to improve the estimation accuracy of models basing on vegetation indices at single-, two- and three-growth stages were tested. With the introduction of florescence spectral information, the single-growth VI-based model produced the most obvious improvement in estimation accuracy, with the coefficient of determination (R2) increasing from 0.748 to 0.799, and the mean absolute percentage error (MAPE) and the root mean squared error (RMSE) decreasing by 11.8% and 10.7%, respectively. Optimized by flowering information, the two-growth stage VIs-based model gave the best performance (R2 = 0.869, MAPE = 3.98%, RMSE = 396.02 kg/ha). These results showed that introducing florescence spectral information at the flowering stage into conventional VIs-based yield estimation models is helpful in improving rice yield estimation accuracy. The usefulness of florescence spectral information for yield estimation provides a new idea for the further development and improvement of the crop yield estimation method.