Image classification with deep learning in the presence of noisy labels: A survey

Fernández-Beltrán

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

et al. 2021

Recently, many deep learning-based methods have been developed for solving remote sensing (RS) scene classification or retrieval tasks. Most of the adopted loss functions for training these models require accurate annotations. However, the presence of noise in such annotations (also known as label noise) cannot be avoided in large-scale RS benchmark archives, resulting from geo-location/registration errors, land-cover changes, and diverse knowledge background of annotators. To overcome the influence of noisy labels on the learning process of deep models, we propose a new loss function called noise-tolerant deep neighborhood embedding which can accurately encode the semantic relationships among RS scenes. Specifically, we target at maximizing the leave-one-out K-NN score for uncovering the inherent neighborhood structure among the images in feature space. Moreover, we down-weight the contribution of potential noisy images by learning their localized structure and pruning the images with low leaveone-out K-NN scores. Based on our newly proposed loss function, classwise features can be more robustly discriminated. Our experiments, conducted on two benchmark RS datasets, validate the effectiveness of the proposed approach on three different RS scene interpretation tasks, including classification, clustering, and retrieval. The codes of this article will be publicly available from https://github.com/jiankang1991.

Section: B Robust Loss Function For Deep Learningmentioning

confidence: 99%

Noise-Tolerant Deep Neighborhood Embedding for Remotely Sensed Images With Label Noise

Fernández-Beltrán

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

et al. 2021

“…The issue could be largely leveraged by using a curriculum strategy (Bengio et al, 2009). Finally, noisy annotations are a typical challenge in machine learning (Natarajan et al, 2013), particularly in image classification and segmentation (Frenay and Verleysen, 2014;Algan and Ulusoy, 2019). Some noise-tolerant versions of CNNs have been developed (Lu et al, 2017; and have achieved various degrees of success in public datasets such as Pascal VOC (Everingham et al, 2010) and CIFAR-10 (Krizhevsky and Hinton, 2009).…”

Section: Abnormal Annotationmentioning

confidence: 99%

Boosting Multilabel Semantic Segmentation for Somata and Vessels in Mouse Brain

Tao

et al. 2021

Front. Neurosci.

Deep convolutional neural networks (DCNNs) are widely utilized for the semantic segmentation of dense nerve tissues from light and electron microscopy (EM) image data; the goal of this technique is to achieve efficient and accurate three-dimensional reconstruction of the vasculature and neural networks in the brain. The success of these tasks heavily depends on the amount, and especially the quality, of the human-annotated labels fed into DCNNs. However, it is often difficult to acquire the gold standard of human-annotated labels for dense nerve tissues; human annotations inevitably contain discrepancies or even errors, which substantially impact the performance of DCNNs. Thus, a novel boosting framework consisting of a DCNN for multilabel semantic segmentation with a customized Dice-logarithmic loss function, a fusion module combining the annotated labels and the corresponding predictions from the DCNN, and a boosting algorithm to sequentially update the sample weights during network training iterations was proposed to systematically improve the quality of the annotated labels; this framework eventually resulted in improved segmentation task performance. The microoptical sectioning tomography (MOST) dataset was then employed to assess the effectiveness of the proposed framework. The result indicated that the framework, even trained with a dataset including some poor-quality human-annotated labels, achieved state-of-the-art performance in the segmentation of somata and vessels in the mouse brain. Thus, the proposed technique of artificial intelligence could advance neuroscience research.

“…1) The current image processing method cannot make pixel-wise ground truth, and manual calibration costs a lot of time, which is unfeasible for subsequent experiments and industrial applications. 2) Different from the traditional threshold and regression algorithm, owing to the gradient back propagation and batch processing mechanism, deep neural network shows great robustness in noisy learning (the process of learning from a dataset containing partial error samples is usually called noisy learning), and has been verified in many experiments [49]. A total of 2,167 images were collected in the experiment.…”

Section: Dataset Establishment and Data Augmentationmentioning

confidence: 99%

Recognition of Multiscale Dense Gel Filament-Droplet Field in Digital Holography With Mo-U-Net

et al. 2021

Accurate particle detection is a common challenge in particle field characterization with digital holography, especially for gel secondary breakup with dense complex particles and filaments of multi-scale and strong background noises. This study proposes a deep learning method called Mo-U-net which is adapted from the combination of U-net and Mobilenetv2, and demostrates its application to segment the dense filament-droplet field of gel drop. Specially, a pruning method is applied on the Mo-U-net, which cuts off about two-thirds of its deep layers to save its training time while remaining a high segmentation accuracy. The performances of the segmentation are quantitatively evaluated by three indices, the positive intersection over union (PIOU), the average square symmetric boundary distance (ASBD) and the diameter-based prediction statistics (DBPS). The experimental results show that the area prediction accuracy (PIOU) of Mo-U-net reaches 83.3%, which is about 5% higher than that of adaptive-threshold method (ATM). The boundary prediction error (ASBD) of Mo-U-net is only about one pixel-wise length, which is one third of that of ATM. And Mo-U-net also shares a coherent size distribution (DBPS) prediction of droplet diameters with the reality. These results demonstrate the high accuracy of Mo-U-net in dense filament-droplet field recognition and its capability of providing accurate statistical data in a variety of holographic particle diagnostics. Public model address: https://github.com/Wu-Tong-Hearted/Recognition-of-multiscale-dense-gel-filament-droplet-field-in-digital-holography-with-Mo-U-net.