查看更多>>摘要：Tomato disease control is of great significance to ensure crop production and tomato disease classification study is an essential tool for doing so. In this paper, we propose a new data augmentation method based on deep threshold multi-feature extraction convolution GAN with Mixed Attention and Markovian Discriminator (MMDGAN) for tomato disease leaf classification. Firstly, in the generator of MMDGAN, a deep threshold multi-feature extraction module was proposed to improve the feature extraction of tomato disease leaves. Then, a mixed attention mechanism combined cross attention module with fused features-highlighting module was proposed to coordinate the overall generation of images. Finally, for the discriminator, Markov discriminator was used to strengthen the similarity judgment of local texture of images. Based on the open datasets PlantVillage, the Frechet Inception Distance (FID) score of healthy tomato leaf image, Leaf Mold, Leaf Curl and Spider Mite generated by MMDGAN were 159.3010, 164.4744, 230.3825 and 254.9866 respectively. Thereafter, a B-ARNet model is trained on synthetic and real images using transfer learning to classify the four categories of tomato diseases. The proposed method achieved an accuracy of 97.12%, with and F1 value of 97.78%. The proposed approach shows its superiority over the existing methodologies.
查看更多>>摘要：Nowadays, recommendation models based on matrix factorization (MF) suffer from the problem of rating sparsity because user-product rating matrix is usually sparse. To address the problem, it is significant to fuse some contextual data or side information on basic MF models. According to this core idea, this paper proposes a modified recommendation model, MFFR (matrix factorization fusing reviews) which recommend products by considering the fusing information on user reviews and user ratings. First, MFFR constructs user-product preference matrix from user reviews by using Latent Dirichlet Allocation (LDA) topic model. Then MFFR predicts ratings and generates personalized top-n recommendation products by using MF model to learn comprehensive latent factors of user-product rating matrix and user-product preference matrix simultaneously. The experimental results of three published datasets demonstrate that our model MFFR can achieve more accurate predicted ratings and hits more correct products of top-n recommendation than the comparative traditional models. MFFR can effectively raise the quality of recommendation, especially in the high level of rating sparsity.
查看更多>>摘要：Cardiovascular disease is a global leading cause of death, and timely monitoring can determine its extent. Clinicians use these diagnostic indicators to make scientific and reasonable decisions. However, when decision-makers (DMs) encounter risks in complex environments, their limited rationality may affect decision behaviors. Therefore, the paper explores a new three-way multi-attribute decision making method based on regret theory (3W-MADM-R), which uses heart disease data to make decisions in fuzzy environments. There are three main steps in developing 3W-MADM-R, i.e., (i) we propose the notion of relative outcome functions and corresponding aggregated regret-based utility functions of each object; (ii) we estimate the conditional probability via an outranked set defined by an outranking relation based on the Preference Ranking Organization Method for Enrichment Evaluation (PROMETHEE II); (iii) we construct three-way decision rules to solve the problems of clustering and ranking of objects in data analysis. In order to demonstrate the usefulness of 3W-MADM-R, we apply it to analyze heart disease data. By comparing with results of other methods, we show the feasibility, stability and superiority of the presented 3W-MADM-R method.
查看更多>>摘要：With the popularity of high-speed rail in China and with faster operation speeds, the safety of railways has also received more attention. This paper proposes a vigilance detection method for high-speed rail drivers based on wireless wearable multi-physiological signals fusion and deep learning. The intended method consists of three components: wireless wearable signal acquisition equipment, physiological signal preprocessing and driver vigilance detection. In the initial stage, a wireless wearable device based on open source brain–computer interfaces was used to collect electroencephalogram, electrocardiogram and electromyogram signals in the high-speed rail simulation environment. Secondly, linear filtering, fast independent component analysis and wavelet filtering are performed on the three kinds of signals, and reasonable slicing is performed to make a dataset. Finally, a convolutional recurrent neural network with channel attention mechanism and memory ability is proposed. Multiple physiological signals from wireless wearable devices are used to train the network. This network improves the ability to recognize the vigilance of drivers and verifies the effectiveness of the combination of squeeze-and-excitation block and long short-term memory with convolutional neural network. Furthermore, the vigilance detection effectiveness was evaluated under different signal combinations; the testing set verified the accuracy of the network as 98.11%. Results prove the feasibility of the high-speed rail driver vigilance detection method based on multiple physiological signals and deep learning, which can help to avoid high-speed rail accidents.
查看更多>>摘要：In the context of the global coronavirus pandemic, different deep learning solutions for infected subject detection using chest X-ray images have been proposed. However, deep learning models usually need large labelled datasets to be effective. Semi-supervised deep learning is an attractive alternative, where unlabelled data is leveraged to improve the overall model's accuracy. However, in real-world usage settings, an unlabelled dataset might present a different distribution than the labelled dataset (i.e. the labelled dataset was sampled from a target clinic and the unlabelled dataset from a source clinic). This results in a distribution mismatch between the unlabelled and labelled datasets. In this work, we assess the impact of the distribution mismatch between the labelled and the unlabelled datasets, for a semi-supervised model trained with chest X-ray images, for COVID-19 detection. Under strong distribution mismatch conditions, we found an accuracy hit of almost 30%, suggesting that the unlabelled dataset distribution has a strong influence in the behaviour of the model. Therefore, we propose a straightforward approach to diminish the impact of such distribution mismatch. Our proposed method uses a density approximation of the feature space. It is built upon the target dataset to filter out the observations in the source unlabelled dataset that might harm the accuracy of the semi-supervised model. It assumes that a small labelled source dataset is available together with a larger source unlabelled dataset. Our proposed method does not require any model training, it is simple and computationally cheap. We compare our proposed method against two popular state of the art out-of-distribution data detectors, which are also cheap and simple to implement. In our tests, our method yielded accuracy gains of up to 32%, when compared to the previous state of the art methods. The good results yielded by our method leads us to argue in favour for a more data-centric approach to improve model's accuracy. Furthermore, the developed method can be used to measure data effectiveness for semi-supervised deep learning model training.
NSTL
Elsevier