Multi-stream Neural Network Research Articles

Improving Mineral Classification Using Multimodal Hyperspectral Point Cloud Data and Multi-Stream Neural Network

In this paper, we leverage multimodal data to classify minerals using a multi-stream neural network. In a previous study on the Tinto dataset, which consisted of a 3D hyperspectral point cloud from the open-pit mine Corta Atalaya in Spain, we successfully identified mineral classes by employing various deep learning models. However, this prior work solely relied on hyperspectral data as input for the deep learning models. In this study, we aim to enhance accuracy by incorporating multimodal data, which includes hyperspectral images, RGB images, and a 3D point cloud. To achieve this, we have adopted a graph-based neural network, known for its efficiency in aggregating local information, based on our past observations where it consistently performed well across different hyperspectral sensors. Subsequently, we constructed a multi-stream neural network tailored to handle multimodality. Additionally, we employed a channel attention module on the hyperspectral stream to fully exploit the spectral information within the hyperspectral data. Through the integration of multimodal data and a multi-stream neural network, we achieved a notable improvement in mineral classification accuracy: 19.2%, 4.4%, and 5.6% on the LWIR, SWIR, and VNIR datasets, respectively.

Detection of Forged Images Using a Combination of Passive Methods Based on Neural Networks

In the current era of social media, the proliferation of images sourced from unreliable origins underscores the pressing need for robust methods to detect forged content, particularly amidst the rapid evolution of image manipulation technologies. Existing literature delineates two primary approaches to image manipulation detection: active and passive. Active techniques intervene preemptively, embedding structures into images to facilitate subsequent authenticity verification, whereas passive methods analyze image content for traces of manipulation. This study presents a novel solution to image manipulation detection by leveraging a multi-stream neural network architecture. Our approach harnesses three convolutional neural networks (CNNs) operating on distinct data streams extracted from the original image. We have developed a solution based on two passive detection methodologies. The system utilizes two separate streams to extract specific data subsets, while a third stream processes the unaltered image. Each net independently processes its respective data stream, capturing diverse facets of the image. The outputs from these nets are then fused through concatenation to ascertain whether the image has undergone manipulation, yielding a comprehensive detection framework surpassing the efficacy of its constituent methods. Our work introduces a unique dataset derived from the fusion of four publicly available datasets, featuring organically manipulated images that closely resemble real-world scenarios. This dataset offers a more authentic representation than other state-of-the-art methods that use algorithmically generated datasets based on image patches. By encompassing genuine manipulation scenarios, our dataset enhances the model’s ability to generalize across varied manipulation techniques, thereby improving its performance in real-world settings. After training, the merged approach obtained an accuracy of 89.59% in the set of validation images, significantly higher than the model trained with only unaltered images, which obtained 78.64%, and the two other models trained using images with a feature selection method applied to enhance inconsistencies that obtained 68.02% for Error-Level Analysis images and 50.70% for the method using Discrete Wavelet Transform. Moreover, our proposed approach exhibits reduced accuracy variance compared to alternative models, underscoring its stability and robustness across diverse datasets. The approach outlined in this work needs to provide information about the specific location or type of tempering, which limits its practical applications.

Viewpoint guided multi-stream neural network for skeleton action recognition

Viewpoint guided multi-stream neural network for skeleton action recognition

Decoding and synthesizing tonal language speech from brain activity.

Recent studies have shown that the feasibility of speech brain-computer interfaces (BCIs) as a clinically valid treatment in helping nontonal language patients with communication disorders restore their speech ability. However, tonal language speech BCI is challenging because additional precise control of laryngeal movements to produce lexical tones is required. Thus, the model should emphasize the features from the tonal-related cortex. Here, we designed a modularized multistream neural network that directly synthesizes tonal language speech from intracranial recordings. The network decoded lexical tones and base syllables independently via parallel streams of neural network modules inspired by neuroscience findings. The speech was synthesized by combining tonal syllable labels with nondiscriminant speech neural activity. Compared to commonly used baseline models, our proposed models achieved higher performance with modest training data and computational costs. These findings raise a potential strategy for approaching tonal language speech restoration.

Hybrid Reinforcement Learning-Based Eco-Driving Strategy for Connected and Automated Vehicles at Signalized Intersections

Taking advantage of both vehicle-to-everything (V2X) communication and automated driving technology, connected and automated vehicles are quickly becoming one of the transformative solutions to many transportation problems. However, in a mixed traffic environment at signalized intersections, it is still a challenging task to improve overall throughput and energy efficiency considering the complexity and uncertainty in the traffic system. In this study, we proposed a hybrid reinforcement learning (HRL) framework which combines the rule-based strategy and the deep reinforcement learning (deep RL) to support connected eco-driving at signalized intersections in mixed traffic. Vision-perceptive methods are integrated with vehicle-to-infrastructure (V2I) communications to achieve higher mobility and energy efficiency in mixed connected traffic. The HRL framework has three components: a rule-based driving manager that operates the collaboration between the rule-based policies and the RL policy; a multi-stream neural network that extracts the hidden features of vision and V2I information; and a deep RL-based policy network that generate both longitudinal and lateral eco-driving actions. In order to evaluate our approach, we developed a Unity-based simulator and designed a mixed-traffic intersection scenario. Moreover, several baselines were implemented to compare with our new design, and numerical experiments were conducted to test the performance of the HRL model. The experiments show that our HRL method can reduce energy consumption by 12.70% and save 11.75% travel time when compared with a state-of-the-art model-based Eco-Driving approach.

A spatiotemporal multi-stream learning framework based on attention mechanism for automatic modulation recognition

Automatic modulation recognition (AMR) plays an essential role in wireless communication systems. Our paper proposes a novel multi-stream neural network (MSNN) to extract the features in parallel from the amplitude, phase, frequency, and raw data of the modulated signal. The framework integrates convolutional neural networks (CNN) and bidirectional gated recurrent units (Bi-GRU) to extract features more effectively from the spatial and temporal characteristics with the assistance of two different attention mechanisms, convolutional block attention module (CBAM) and multi-head self-attention (MHSA). Simulation experiments show that the performance of our proposed algorithm is better than that of other state-of-the-art (SOTA) recognition algorithms on four widely used DeepSig datasets. The recognition accuracy of our proposed model exceeds 99% (18 dB), 93% (16 dB), 95% (20 dB), and 97% (20 dB) respectively on the datasets RML2016.04C, RML2016.10A, RML2016.10B, and RML2018.01A.

Word-Level Sign Language Recognition with Multi-Stream Neural Networks Focusing on Local Regions and Skeletal Information

Word-Level Sign Language Recognition with Multi-Stream Neural Networks Focusing on Local Regions and Skeletal Information

Single-image De-raining with a Connected Multi-stream Neural Network

Single-image De-raining with a Connected Multi-stream Neural Network

Deep learning with attention supervision for automated motion artefact detection in quality control of cardiac T1-mapping.

Cardiac magnetic resonance quantitative T1-mapping is increasingly used for advanced myocardial tissue characterisation. However, cardiac or respiratory motion can significantly affect the diagnostic utility of T1-maps, and thus motion artefact detection is critical for quality control and clinically-robust T1 measurements. Manual quality control of T1-maps may provide reassurance, but is laborious and prone to error. We present a deep learning approach with attention supervision for automated motion artefact detection in quality control of cardiac T1-mapping. Firstly, we customised a multi-stream Convolutional Neural Network (CNN) image classifier to streamline the process of automatic motion artefact detection. Secondly, we imposed attention supervision to guide the CNN to focus on targeted myocardial segments. Thirdly, when there was disagreement between the human operator and machine, a second human validator reviewed and rescored the cases for adjudication and to identify the source of disagreement. The multi-stream neural networks demonstrated 89.8% agreement, 87.4% ROC-AUC on motion artefact detection with the human operator in the 2568 T1 maps. Trained with additional supervision on attention, agreements and AUC significantly improved to 91.5% and 89.1%, respectively (p < 0.001). Rescoring of disagreed cases by the second human validator revealed that human operator error was the primary cause of disagreement. Deep learning with attention supervision provides a quick and high-quality assurance of clinical images, and outperforms human operators.

Multi-stream neural network fused with local information and global information for HOI detection

Human-Object Interaction (HOI) Detection is a new genre of human-centric visual relationship detection task, which is significant to deep understanding of visual scenes. Due to the complexity of the visual scene in the image, HOI detection is still a challenging task, the most critical part of which is feature extraction and representation. Some existing approaches rely solely on local region information for HOI detection without using global contextual information, but global contextual information contributes to this task in some HOI categories. Other approaches incorporate global contextual information for HOI detection while losing local region information. In this work, we propose a multi-stream neural network architecture composed of three special module that employs both local region information and global contextual information for HOI detection. This model can detect not only the HOI categories based on local region information but also on global contextual information. Our model more fully considers all HOI categories in the dataset. Compared with other existing approaches, the proposed model shows improved performance on V-COCO and HICO-DET benchmark datasets, especially when predicting rare HOI categories.

Learning Representations From Skeletal Self-Similarities for Cross-View Action Recognition

Existing research attention in vision-based action recognition is generally paid on recognizing actions from the same views seen in the training data. One of the big challenges in action recognition lies in the large variations of action representations as actions are captured from totally different viewpoints. This paper addresses this problem by learning view-invariant representations from skeletal self-similarities of varying scales with a very light multi-stream neural network (MSNN). As human skeletons have been proved to be an effective feature modality used for action recognition and are easy to obtain, we first create a view-invariant action description by formulating skeletal self-similarities at each frame as an image (SSI), which can show a high structural stability under view changes. Accordingly, a MSNN is designed based on 3D CNN and LSTM units to learn representations from SSIs of multiple scales, where the scheme of multiple scales provides our method with a good robustness to view changes. In addition, we integrate the computation of SSIs into the MSNN by wrapping it as a custom learnable layer thanks to its simplicity, instead of normalizing and transforming skeletons using a hand-crafted preprocessing. Extensive experimental evaluations on three challenging cross-view datasets demonstrate the effectiveness of our proposed method, which achieves superior performance to the state-of-the-art algorithms on cross-view recognition. The source code of this work will be released shortly to facilitate future studies in this field.

Background Knowledge Based Multi-Stream Neural Network for Text Classification

As a foundation and typical task in natural language processing, text classification has been widely applied in many fields. However, as the basis of text classification, most existing corpus are imbalanced and often result in the classifier tending its performance to those categories with more texts. In this paper, we propose a background knowledge based multi-stream neural network to make up for the imbalance or insufficient information caused by the limitations of training corpus. The multi-stream network mainly consists of the basal stream, which retained original sequence information, and background knowledge based streams. Background knowledge is composed of keywords and co-occurred words which are extracted from external corpus. Background knowledge based streams are devoted to realizing supplemental information and reinforce basal stream. To better fuse the features extracted from different streams, early-fusion and two after-fusion strategies are employed. According to the results obtained from both Chinese corpus and English corpus, it is demonstrated that the proposed background knowledge based multi-stream neural network performs well in classification tasks.

Local online learning of coherent information

One of the goals of perception is to learn to respond to coherence across space, time and modality. Here we present an abstract framework for the local online unsupervised learning of this coherent information using multi-stream neural networks. The processing units distinguish between feedforward inputs projected from the environment and the lateral, contextual inputs projected from the processing units of other streams. The contextual inputs are used to guide learning towards coherent cross-stream structure. The goal of all the learning algorithms described is to maximize the predictability between each unit output and its context. Many local cost functions may be applied: e.g. mutual information, relative entropy, squared error and covariance. Theoretical and simulation results indicate that, of these, the covariance rule (1) is the only rule that specifically links and learns only those streams with coherent information, (2) can be robustly approximated by a Hebbian rule, (3) is stable with input noise, no pairwise input correlations, and in the discovery of locally less informative components that are coherent globally. In accordance with the parallel nature of the biological substrate, we also show that all the rules scale up with the number of streams.

Contextually guided unsupervised learning using local multivariate binary processors

We consider the role of contextual guidance in learning and processing within multi-stream neural networks. Earlier work ( Kay and Phillips 1994, 1996; Phillips et al., 1995) showed how the goals of feature discovery and associative learning could be fused within a single objective and made precise using information theory in such a way that local binary processors could extract a single feature that is coherent across streams. In this paper, we consider multi-unit local processors with multivariate binary outputs that enable a greater number of coherent features to be extracted. Using the Ising model, we define a class of information–theoretic objective functions and also local approximations and derive the learning rules in both cases. These rules have similarities to, and differences from, the celebrated BCM rule. Local and global versions of infomax appear as by-products of the general approach, as well as multivariate versions of coherent infomax. Focussing on the more biologically plausible local rules, we describe some computational experiments designed to investigate specific properties of the processors and the general approach. The main conclusions are: (1) the local methodology introduced in the paper has the required functionality. (2) Different units within the multi-unit processors learned to respond to different aspects of their receptive fields. (3) The units within each processor generally produced a distributed code in which the outputs were correlated and which was robust to damage; in the special case where the number of units available was only just sufficient to transmit the relevant information, a form of competitive learning was produced. (4) The contextual connections enabled the information correlated across streams to be extracted and, by improving feature detection with weak or noisy inputs, they played a useful role in short-term processing and in improving generalization. (5) The methodology allows the statistical associations between distributed self-organizing population codes to be learned.