Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Machine Learning-Based Microclimate Model for Indoor Air Temperature and Relative Humidity Prediction in a Swine Building.

In Animals : an open access journal from MDPI

Indoor air temperature (IAT) and indoor relative humidity (IRH) are the prominent microclimatic variables; still, potential contributors that influence the homeostasis of livestock animals reared in closed barns. Further, predicting IAT and IRH encourages farmers to think ahead actively and to prepare the optimum solutions. Therefore, the primary objective of the current literature is to build and investigate extensive performance analysis between popular ML models in practice used for IAT and IRH predictions. Meanwhile, multiple linear regression (MLR), multilayered perceptron (MLP), random forest regression (RFR), decision tree regression (DTR), and support vector regression (SVR) models were utilized for the prediction. This study used accessible factors such as external environmental data to simulate the models. In addition, three different input datasets named S1, S2, and S3 were used to assess the models. From the results, RFR models performed better results in both IAT (R2 = 0.9913; RMSE = 0.476; MAE = 0.3535) and IRH (R2 = 0.9594; RMSE = 2.429; MAE = 1.47) prediction among other models particularly with S3 input datasets. In addition, it has been proven that selecting the right features from the given input data builds supportive conditions under which the expected results are available. Overall, the current study demonstrates a better model among other models to predict IAT and IRH of a naturally ventilated swine building containing animals with fewer input attributes.

Arulmozhi Elanchezhian, Basak Jayanta Kumar, Sihalath Thavisack, Park Jaesung, Kim Hyeon Tae, Moon Byeong Eun

2021-Jan-18

ML models, indoor air temperature, indoor relative humidity, smart farming, swine building microclimate

General General

EnlightenGAN: Deep Light Enhancement without Paired Supervision.

In IEEE transactions on image processing : a publication of the IEEE Signal Processing Society

Deep learning-based methods have achieved remarkable success in image restoration and enhancement, but are they still competitive when there is a lack of paired training data? As one such example, this paper explores the low-light image enhancement problem, where in practice it is extremely challenging to simultaneously take a low-light and a normal-light photo of the same visual scene. We propose a highly effective unsupervised generative adversarial network, dubbed EnlightenGAN, that can be trained without low/normal-light image pairs, yet proves to generalize very well on various real-world test images. Instead of supervising the learning using ground truth data, we propose to regularize the unpaired training using the information extracted from the input itself, and benchmark a series of innovations for the low-light image enhancement problem, including a global-local discriminator structure, a self-regularized perceptual loss fusion, and the attention mechanism. Through extensive experiments, our proposed approach outperforms recent methods under a variety of metrics in terms of visual quality and subjective user study. Thanks to the great flexibility brought by unpaired training, EnlightenGAN is demonstrated to be easily adaptable to enhancing real-world images from various domains. Our codes and pre-trained models are available at: https://github.com/VITA-Group/EnlightenGAN.

Jiang Yifan, Gong Xinyu, Liu Ding, Cheng Yu, Fang Chen, Shen Xiaohui, Yang Jianchao, Zhou Pan, Wang Zhangyang

2021-Jan-22

General General

Interpretable Detail-Fidelity Attention Network for Single Image Super-Resolution.

In IEEE transactions on image processing : a publication of the IEEE Signal Processing Society

Benefiting from the strong capabilities of deep CNNs for feature representation and nonlinear mapping, deep-learning-based methods have achieved excellent performance in single image super-resolution. However, most existing SR methods depend on the high capacity of networks that are initially designed for visual recognition, and rarely consider the initial intention of super-resolution for detail fidelity. To pursue this intention, there are two challenging issues that must be solved: (1) learning appropriate operators which is adaptive to the diverse characteristics of smoothes and details; (2) improving the ability of the model to preserve low-frequency smoothes and reconstruct high-frequency details. To solve these problems, we propose a purposeful and interpretable detail-fidelity attention network to progressively process these smoothes and details in a divide-and-conquer manner, which is a novel and specific prospect of image super-resolution for the purpose of improving detail fidelity. This proposed method updates the concept of blindly designing or using deep CNNs architectures for only feature representation in local receptive fields. In particular, we propose a Hessian filtering for interpretable high-profile feature representation for detail inference, along with a dilated encoder-decoder and a distribution alignment cell to improve the inferred Hessian features in a morphological manner and statistical manner respectively. Extensive experiments demonstrate that the proposed method achieves superior performance compared to the state-of-the-art methods both quantitatively and qualitatively. The code is available at https://github.com/YuanfeiHuang/DeFiAN.

Huang Yuanfei, Li Jie, Gao Xinbo, Hu Yanting, Lu Wen

2021-Jan-22

General General

Part-Object Relational Visual Saliency.

In IEEE transactions on pattern analysis and machine intelligence ; h5-index 127.0

Recent years have witnessed a big leap in automatic visual saliency detection attributed to advances in deep learning, especially Convolutional Neural Networks (CNNs). However, inferring the saliency of each image part separately, as was adopted by most CNNs methods, inevitably leads to an incomplete segmentation of the salient object. In this paper, we describe how to use the property of part-object relations endowed by the Capsule Network (CapsNet) to solve the problems that fundamentally hinge on relational inference for visual saliency detection. Concretely, we put in place a two-stream strategy, termed Two-Stream Part-Object RelaTional Network (TSPORTNet), to implement CapsNet, aiming to reduce both the network complexity and the possible redundancy during capsule routing. Additionally, taking into account the correlations of capsule types from the preceding training images, a correlation-aware capsule routing algorithm is developed for more accurate capsule assignments at the training stage, which also speeds up the training dramatically. By exploring part-object relationships, TSPORTNet produces a capsule wholeness map, which in turn aids multi-level features in generating the final saliency map. Experimental results on five widely-used benchmarks show that our framework consistently achieves state-of-the-art performance. The code can be found on https://github.com/liuyi1989/TSPORTNet.

Liu Yi, Zhang Dingwen, Zhang Qiang, Han Jungong

2021-Jan-22

General General

Hemocompatibility Evaluation of Biomaterials-The Crucial Impact of Analyzed Area.

In ACS biomaterials science & engineering ; h5-index 39.0

The hemocompatibility of blood-contacting medical devices remains one of the major challenges in medical device development. A common tool for the analysis of adherent and activated platelets on materials following in vitro tests is microscopy. Currently, most researchers develop their own routines, resulting in numerous different methods that are applied. The majority of those (semi-)manual methods analyze only a very small fraction of the material surface (<1%), which neglects the inhomogeneity of platelet distribution and makes results hardly comparable. Within this study, we examined the relation between the fraction of analyzed sample area and the platelet adhesion result. By means of image segmentation and machine learning algorithms, 103 100 microscopy images were analyzed automatically. We discovered a crucial impact of the analyzed surface fraction and thus a misrepresentation of a surface's platelet adhesion unless up to 40% of the sample surface is analyzed. These findings underline the necessity of standardization in the field of in vitro hemocompatibility tests and analyses in particular and provide a first basis to make future tests more reliable and comparable.

Clauser Johanna C, Maas Judith, Arens Jutta, Schmitz-Rode Thomas, Steinseifer Ulrich, Berkels Benjamin

2021-Jan-22

automation, fluorescence microscopy, in vitro testing, platelet analysis, standardization

General General

Mismatched response predicts behavioral speech discrimination outcomes in infants with hearing loss and normal hearing.

In Infancy : the official journal of the International Society on Infant Studies

Children with hearing loss (HL) remain at risk for poorer language abilities than normal hearing (NH) children despite targeted interventions; reasons for these differences remain unclear. In NH children, research suggests speech discrimination is related to language outcomes, yet we know little about it in children with HL under the age of 2 years. We utilized a vowel contrast, /a-i/, and a consonant-vowel contrast, /ba-da/, to examine speech discrimination in 47 NH infants and 40 infants with HL. At Mean age =3 months, EEG recorded from 11 scalp electrodes was used to compute the time-frequency mismatched response (TF-MMRSE ) to the contrasts; at Mean age =9 months, behavioral discrimination was assessed using a head turn task. A machine learning (ML) classifier was used to predict behavioral discrimination when given an arbitrary TF-MMRSE as input, achieving accuracies of 73% for exact classification and 92% for classification within a distance of one class. Linear fits revealed a robust relationship regardless of hearing status or speech contrast. TF-MMRSE responses in the delta (1-3.5 Hz), theta (3.5-8 Hz), and alpha (8-12 Hz) bands explained the most variance in behavioral task performance. Our findings demonstrate the feasibility of using TF-MMRSE to predict later behavioral speech discrimination.

Uhler Kristin, Hunter Sharon, Gilley Phillip M

2021-Jan-22