Lightweight Deep Network Research Articles

Lightweight Multilevel Feature Fusion Network for Hyperspectral Image Classification

Hyperspectral images (HSIs), acquired as a 3D data set, contain spectral and spatial information that is important for ground–object recognition. A 3D convolutional neural network (3DCNN) could therefore be more suitable than a 2D one for extracting multiscale neighborhood information in the spectral and spatial domains simultaneously, if it is not restrained by mass parameters and computation cost. In this paper, we propose a novel lightweight multilevel feature fusion network (LMFN) that can achieve satisfactory HSI classification with fewer parameters and a lower computational burden. The LMFN decouples spectral–spatial feature extraction into two modules: point-wise 3D convolution to learn correlations between adjacent bands with no spatial perception, and depth-wise convolution to obtain local texture features while the spectral receptive field remains unchanged. Then, a target-guided fusion mechanism (TFM) is introduced to achieve multilevel spectral–spatial feature fusion between the two modules. More specifically, multiscale spectral features are endowed with spatial long-range dependency, which is quantified by central target pixel-guided similarity measurement. Subsequently, the results obtained from shallow to deep layers are added, respectively, to the spatial modules, in an orderly manner. The TFM block can enhance adjacent spectral correction and focus on pixels that actively boost the target classification accuracy, while performing multiscale feature fusion. Experimental results across three benchmark HSI data sets indicate that our proposed LMFN has competitive advantages, in terms of both classification accuracy and lightweight deep network architecture engineering. More importantly, compared to state-of-the-art methods, the LMFN presents better robustness and generalization.

Multipath Lightweight Deep Network Using Randomly Selected Dilated Convolution.

Robot vision is an essential research field that enables machines to perform various tasks by classifying/detecting/segmenting objects as humans do. The classification accuracy of machine learning algorithms already exceeds that of a well-trained human, and the results are rather saturated. Hence, in recent years, many studies have been conducted in the direction of reducing the weight of the model and applying it to mobile devices. For this purpose, we propose a multipath lightweight deep network using randomly selected dilated convolutions. The proposed network consists of two sets of multipath networks (minimum 2, maximum 8), where the output feature maps of one path are concatenated with the input feature maps of the other path so that the features are reusable and abundant. We also replace the standard convolution of each path with a randomly selected dilated convolution, which has the effect of increasing the receptive field. The proposed network lowers the number of floating point operations (FLOPs) and parameters by more than 50% and the classification error by 0.8% as compared to the state-of-the-art. We show that the proposed network is efficient.

A Lightweight Deep Network for Defect Detection of Insert Molding Based on X-ray Imaging.

Aiming at the abnormality detection of industrial insert molding processes, a lightweight but effective deep network is developed based on X-ray images in this study. The captured digital radiography (DR) images are firstly fast guide filtered, and then a multi-task detection dataset is constructed using an overlap slice in order to improve the detection of tiny targets. The proposed network is extended from the one-stage target detection method of yolov5 to be applicable to DR defect detection. We adopt the embedded Ghost module to replace the standard convolution to further lighten the model for industrial implementation, and use the transformer module for spatial multi-headed attentional feature extraction to perform improvement on the network for the DR image defect detection. The performance of the proposed method is evaluated by consistent experiments with peer networks, including the classical two-stage method and the newest yolo series. Our method achieves a mAP of 93.6%, which exceeds the second best by 3%, with robustness sufficient to cope with luminance variations and blurred noise, and is more lightweight. We further conducted ablation experiments based on the proposed method to validate the 32% model size reduction owing to the Ghost module and the detection performance enhancing effect of other key modules. Finally, the usability of the proposed method is discussed, including an analysis of the common causes of the missed shots and suggestions for modification. Our proposed method contributes a good reference solution for the inspection of the insert molding process.

Nighttime object detection system with lightweight deep network for internet of vehicles

Nighttime object detection system with lightweight deep network for internet of vehicles

A Lightweight Deep Network for Efficient CSI Feedback in Massive MIMO Systems

To fully exploit the advantages of massive multiple-input multiple-output (m-MIMO), accurate channel state information (CSI) is required at the transmitter. However, excessive CSI feedback for large antenna arrays is inefficient and thus undesirable in practical applications. By exploiting the inherent correlation characteristics of complex-valued channel responses in the angular-delay domain, we propose a novel neural network (NN) architecture, namely ENet, for CSI compression and feedback in m-MIMO. Even if the ENet processes the real and imaginary parts of the CSI values separately, its special structure enables the network trained for the real part only to be reused for the imaginary part. The proposed ENet shows enhanced performance with the network size reduced by nearly an order of magnitude compared to the existing NN-based solutions. Experimental results verify the effectiveness of the proposed ENet.

Learning to Enhance Low-Light Image via Zero-Reference Deep Curve Estimation.

This paper presents a novel method, Zero-Reference Deep Curve Estimation (Zero-DCE), which formulates light enhancement as a task of image-specific curve estimation with a deep network. Our method trains a lightweight deep network, DCE-Net, to estimate pixel-wise and high-order curves for dynamic range adjustment of a given image. The curve estimation is specially designed, considering pixel value range, monotonicity, and differentiability. Zero-DCE is appealing in its relaxed assumption on reference images, i.e., it does not require any paired or even unpaired data during training. This is achieved through a set of carefully formulated non-reference loss functions, which implicitly measure the enhancement quality and drive the learning of the network. Despite its simplicity, we show that it generalizes well to diverse lighting conditions. Our method is efficient as image enhancement can be achieved by an intuitive and simple nonlinear curve mapping. We further present an accelerated and light version of Zero-DCE, called Zero-DCE++, that takes advantage of a tiny network with just 10K parameters. Zero-DCE++ has a fast inference speed (1000/11 FPS on a single GPU/CPU for an image of size 1200×900×3) while keeping the enhancement performance of Zero-DCE. Extensive experiments on various benchmarks demonstrate the advantages of our method over state-of-the-art methods qualitatively and quantitatively. Furthermore, the potential benefits of our method to face detection in the dark are discussed. The source code is made publicly available at https://li-chongyi.github.io/Proj_Zero-DCE++.html.

Semantic Segmentation of the Eye With a Lightweight Deep Network and Shape Correction

In this paper, we present a multi-class eye segmentation method that can run the hardware limitations for real-time inference. Our approach includes three major stages: get a grayscale image from the input, segment three distinct eye region with a deep network, and remove incorrect areas with heuristic filters. Our model based on the encoder decoder structure with the key is the depthwise convolution operation to reduce the computation cost. We experiment on OpenEDS, a large scale dataset of eye images captured by a head-mounted display with two synchronized eye facing cameras. We achieved the mean intersection over union (mIoU) of 94.85% with a model of size 0.4 megabytes. The source code are available https://github.com/th2l/Eye_VR_Segmentation

Cross Model Deep Learning Scheme for Automatic Modulation Classification

Deep Neural Networks (DNNs) have achieved remarkable accuracy improvements for automatic modulation classification. However, the employed networks often have millions of parameters and need very high computation, which makes it difficult to deploy these models on portable devices with limited resources. We propose a cross model deep learning scheme to build a lightweight deep network for accurate modulation classification. Firstly, a large Hybrid DNN (HDNN) that is composed of convolutional and recurrent layers is constructed and trained for automatic and accurate classification of signals. Then we build a smaller Layered Resnet Network (LRN) with shallow layers and few nodes. The HDNN and LRN are taken as a Teacher Model (TM) and a Student Model (SM) respectively. Finally, a knowledge distillation method is proposed to guide the learning of the SM, by formulating a teaching loss from the prediction of the TM to train the SM. The performances of the proposed HDNN and LRN are investigated on the public RadioML2016.10a and RadioML2016.10b data sets. The experimental results show that the trained HDNN presents state-of-the-art classification results and the LRN trained in this scheme takes only about a sixth of the HDNN's inference time and consumes only 472.3KB for storage, with a slight accuracy decrease compared with the large HDNN.

Recognition and Clinical Diagnosis of Cervical Cancer Cells Based on our Improved Lightweight Deep Network for Pathological Image.

Accurate recognition of cervical cancer cells is of great significance to clinical diagnosis, but these existing algorithms are designed by low-level manual feature, and their performance improvements are limited an improved algorithm based on residual neural network is proposed to improve the accuracy of diagnosis. Firstly, momentum parameters are introduced into the training model; secondly, by changing the number of training samples, the recognition rate of the algorithm can be improved. Therefore, aiming at the task of object recognition under resource constrained condition, we optimize the design method of the network structure such as convolution operation, model parameter compression and enhancement of feature expression depth, and design and implement the lightweight network model structure for embedded platform. Our proposed deep network model can reduce the parameters of the model and the resources needed for operation under the condition of guaranteeing the precision. The experimental results show that the lightweight deep model has better performance than that of the existing comparison models, and it can achieve the model accuracy of 94.1% under the condition that the model with fewer parameters on cervical cells data set.

Lightweight deep network for traffic sign classification

Deeper neural networks have achieved great results in the field of computer vision and have been successfully applied to tasks such as traffic sign recognition. However, as traffic sign recognition systems are often deployed in resource-constrained environments, it is critical for the network design to be slim and accurate in these instances. Accordingly, in this paper, we propose two novel lightweight networks that can obtain higher recognition precision while preserving less trainable parameters in the models. Knowledge distillation transfers the knowledge in a trained model, called the teacher network, to a smaller model, called the student network. Moreover, to improve the accuracy of traffic sign recognition, we also implement a new module in our teacher network that combines two streams of feature channels with dense connectivity. To enable easy deployment on mobile devices, our student network is a simple end-to-end architecture containing five convolutional layers and a fully connected layer. Furthermore, by referring to the values of batch normalization (BN) scaling factors towards zero to identify insignificant channels, we prune redundant channels from the student network, yielding a compact model with accuracy comparable to that of more complex models. Our teacher network exhibited an accuracy rate of 93.16% when trained and tested on the CIFAR-10 general dataset. Using the knowledge of our teacher network, we train the student network on the GTSRB and BTSC traffic sign datasets. Thus, our student model uses only 0.8 million parameters while still achieving accuracy of 99.61% and 99.13% respectively on both datasets. All experimental results show that our lightweight networks can be useful when deploying deep convolutional neural networks (CNNs) on mobile embedded devices.

Fast Deep Convolutional Face Detection in the Wild Exploiting Hard Sample Mining

Face detection constitutes a key visual information analysis task in Machine Learning. The rise of Big Data has resulted in the accumulation of a massive volume of visual data which requires proper and fast analysis. Deep Learning methods are powerful approaches towards this task as training with large amounts of data exhibiting high variability has been shown to significantly enhance their effectiveness, but often requires expensive computations and leads to models of high complexity. When the objective is to analyze visual content in massive datasets, the complexity of the model becomes crucial to the success of the model. In this paper, a lightweight deep Convolutional Neural Network (CNN) is introduced for the purpose of face detection, designed with a view to minimize training and testing time, and outperforms previously published deep convolutional networks in this task, in terms of both effectiveness and efficiency. To train this lightweight deep network without compromising its efficiency, a new training method of progressive positive and hard negative sample mining is introduced and shown to drastically improve training speed and accuracy. Additionally, a separate deep network was trained to detect individual facial features and a model that combines the outputs of the two networks was created and evaluated. Both methods are capable of detecting faces under severe occlusion and unconstrained pose variation and meet the difficulties of large scale real-world, real-time face detection, and are suitable for deployment even in mobile environments such as Unmanned Aerial Vehicles (UAVs).