Depthwise Separable Convolutional Neural Network Research Articles

Recognition of Crop Diseases Based on Depthwise Separable Convolution in Edge Computing.

The original pattern recognition and classification of crop diseases needs to collect a large amount of data in the field and send them next to a computer server through the network for recognition and classification. This method usually takes a long time, is expensive, and is difficult to carry out for timely monitoring of crop diseases, causing delays to diagnosis and treatment. With the emergence of edge computing, one can attempt to deploy the pattern recognition algorithm to the farmland environment and monitor the growth of crops promptly. However, due to the limited resources of the edge device, the original deep recognition model is challenging to apply. Due to this, in this article, a recognition model based on a depthwise separable convolutional neural network (DSCNN) is proposed, which operation particularities include a significant reduction in the number of parameters and the amount of computation, making the proposed design well suited for the edge. To show its effectiveness, simulation results are compared with the main convolution neural network (CNN) models LeNet and Visual Geometry Group Network (VGGNet) and show that, based on high recognition accuracy, the recognition time of the proposed model is reduced by 80.9% and 94.4%, respectively. Given its fast recognition speed and high recognition accuracy, the model is suitable for the real-time monitoring and recognition of crop diseases by provisioning remote embedded equipment and deploying the proposed model using edge computing.

A diagnosis method based on depthwise separable convolutional neural network for the attachment on the blade of marine current turbine

To diagnose the attachment of marine current turbine, this article proposes a method based on convolutional neural network and the concepts of depthwise separable convolution to achieve feature extraction. The method consists of three steps: data preprocessing, feature extraction and fault diagnosis. This method can diagnose the fault degree of blade imbalance and uniform attachment in underwater environment with strong currents and complex spatiotemporal variability. It can extract distinct image feature in harsh marine environments by using a convolutional neural network. In addition, this method is robust for the recognition of blurred pictures under high-speed rotation.

A depthwise separable convolutional neural network for keyword spotting on an embedded system

A keyword spotting algorithm implemented on an embedded system using a depthwise separable convolutional neural network classifier is reported. The proposed system was derived from a high-complexity system with the goal to reduce complexity and to increase efficiency. In order to meet the requirements set by hardware resource constraints, a limited hyper-parameter grid search was performed, which showed that network complexity could be drastically reduced with little effect on classification accuracy. It was furthermore found that quantization of pre-trained networks using mixed and dynamic fixed point principles could reduce the memory footprint and computational requirements without lowering classification accuracy. Data augmentation techniques were used to increase network robustness in unseen acoustic conditions by mixing training data with realistic noise recordings. Finally, the system’s ability to detect keywords in a continuous audio stream was successfully demonstrated.

OSVFuseNet: Online Signature Verification by feature fusion and depth-wise separable convolution based deep learning

Online Signature Verification (OSV) techniques have been deployed in production systems for decades, yet training the model for efficient classification of the test signature from fewer training signature samples is still an open challenge. The advancements in Convolutional Neural Networks (CNNs) enormously boosted the effectiveness of OSV systems. However, learning subtle and discriminating representations from few training samples to classify the genuineness of test signature has not been explored fully. In this paper, a Convolution Autoencoder (CAE) is used to obtain high-level feature representations from the input signature and these high level features are fused with handcrafted features to constitute a hybrid feature set. The hybrid set of features is presented as an input to an Online Signature Verification framework made up of Depth-wise Separable Convolutional Neural Network (DWSCNN). DWSCNN effectively learn deep feature representations with fewer training samples and parameters than traditional CNNs resulting in a light weight OSV framework. Thorough experimental analysis on three benchmark datasets MCYT-100 (DB1), SVC-2004-Task2 and SUSIG-Visual corpus confirm that the proposed hybrid fusion of feature set and DWSCNN based OSV framework achieve higher classification accuracies and outperforms many contemporary and state-of-the art OSV models.

Fast and Accurate Algorithm for ECG Authentication Using Residual Depthwise Separable Convolutional Neural Networks

The electrocardiogram (ECG) is relatively easy to acquire and has been used for reliable biometric authentication. Despite growing interest in ECG authentication, there are still two main problems that need to be tackled, i.e., the accuracy and processing speed. Therefore, this paper proposed a fast and accurate ECG authentication utilizing only two stages, i.e., ECG beat detection and classification. By minimizing time-consuming ECG signal pre-processing and feature extraction, our proposed two-stage algorithm can authenticate the ECG signal around 660 μs. Hamilton’s method was used for ECG beat detection, while the Residual Depthwise Separable Convolutional Neural Network (RDSCNN) algorithm was used for classification. It was found that between six and eight ECG beats were required for authentication of different databases. Results showed that our proposed algorithm achieved 100% accuracy when evaluated with 48 patients in the MIT-BIH database and 90 people in the ECG ID database. These results showed that our proposed algorithm outperformed other state-of-the-art methods.

Dense connection and depthwise separable convolution based CNN for polarimetric SAR image classification

Convolution neural networks (CNN) have achieved great success in natural image processing where large amounts of training data are available. However, for the polarimetric synthetic aperture radar (PolSAR) image classification problem, the number of labeled training samples is typically limited. To improve the performance of CNN on limited training data, we propose a new network, the densely connected and depthwise separable convolutional neural network (DSNet). According to characteristics of PolSAR data, DSNet uses depthwise separable convolution to replace standard convolution, to independently extract features over each channel in PolSAR images. DSNet also introduces dense connections to directly connect non-adjacent layers. With the depthwise separable convolution and dense connections, DSNet can avoid extracting redundant features, reuse the hierarchical feature maps of PolSAR images and reduce the number of training parameters. Compared with normal CNN, DSNet is more lightweight and its training parameters decrease to less than 1/9. We compare DSNet against several popular algorithms on three different data sets, and show that DSNet achieves better results while using less training samples.

Attention Receptive Pyramid Network for Ship Detection in SAR Images

With the development of deep learning (DL) and synthetic aperture radar (SAR) imaging techniques, SAR automatic target recognition has come to a breakthrough. Numerous algorithms have been proposed and competitive results have been achieved in detecting different targets. However, due to the influence of various sizes and complex background of ships, detecting multiscale ships in SAR images is still challenging. To solve the problems, a novel network, called attention receptive pyramid network (ARPN), is proposed in this article. ARPN is a two-stage detector and designed to improve the performance of detecting multiscale ships in SAR images by enhancing the relationships among nonlocal features and refining information at different feature maps. Specifically, receptive fields block (RFB) and convolutional block attention module (CBAM) are employed and combined reasonably in attention receptive block to build a top-down fine-grained feature pyramid. RFB, composed of several branches of convolutional layers with specifically asymmetric kernel sizes and various dilation rates, is used for grabbing features of ships with large aspect ratios and enhancing local features with their global dependences. CBAM, which consists of channel and spatial attention mechanisms, is utilized to boost significant information and suppress interference caused by surroundings. To evaluate the effectiveness of ARPN, experiments are conducted on SAR Ship Detection Dataset and two large-scene SAR images. The detection results illustrate that competitive performance has been achieved by our method in comparison with several CNN-based algorithms, e.g., Faster-RCNN, RetinaNet, feature pyramid network, YOLOv3, Dense Attention Pyramid Network, Depth-wise Separable Convolutional Neural Network, High-Resolution Ship Detection Network, and Squeeze and Excitation Rank Faster-RCNN.

Depthwise Separable Convolution Neural Network for High-Speed SAR Ship Detection

As an active microwave imaging sensor for the high-resolution earth observation, synthetic aperture radar (SAR) has been extensively applied in military, agriculture, geology, ecology, oceanography, etc., due to its prominent advantages of all-weather and all-time working capacity. Especially, in the marine field, SAR can provide numerous high-quality services for fishery management, traffic control, sea-ice monitoring, marine environmental protection, etc. Among them, ship detection in SAR images has attracted more and more attention on account of the urgent requirements of maritime rescue and military strategy formulation. Nowadays, most researches are focusing on improving the ship detection accuracy, while the detection speed is frequently neglected, regardless of traditional feature extraction methods or modern deep learning (DL) methods. However, the high-speed SAR ship detection is of great practical value, because it can provide real-time maritime disaster rescue and emergency military planning. Therefore, in order to address this problem, we proposed a novel high-speed SAR ship detection approach by mainly using depthwise separable convolution neural network (DS-CNN). In this approach, we integrated multi-scale detection mechanism, concatenation mechanism and anchor box mechanism to establish a brand-new light-weight network architecture for the high-speed SAR ship detection. We used DS-CNN, which consists of a depthwise convolution (D-Conv2D) and a pointwise convolution (P-Conv2D), to substitute for the conventional convolution neural network (C-CNN). In this way, the number of network parameters gets obviously decreased, and the ship detection speed gets dramatically improved. We experimented on an open SAR ship detection dataset (SSDD) to validate the correctness and feasibility of the proposed method. To verify the strong migration capacity of our method, we also carried out actual ship detection on a wide-region large-size Sentinel-1 SAR image. Ultimately, under the same hardware platform with NVIDIA RTX2080Ti GPU, the experimental results indicated that the ship detection speed of our proposed method is faster than other methods, meanwhile the detection accuracy is only lightly sacrificed compared with the state-of-art object detectors. Our method has great application value in real-time maritime disaster rescue and emergency military planning.

A deep separable neural network for human tissue identification in three-dimensional optical coherence tomography images

This research proposes a dilated depthwise separable network for human tissue identification using three-dimensional (3D) optical coherence tomography (OCT) images. Automatic human tissue identification has made it possible for fast pathological tissue analyses, detecting tissue changes over time, and efficiently making a precise therapy treatment plan. 3D medical image classification is a challenging task because of the indistinct tissue characteristics and computational efficiency. To address the challenge, a deep dilated depthwise separable convolutional neural network is proposed in this research. A depthwise separable architecture is introduced to improve parameter utilization efficiency. Dilated convolutions are applied to systematically aggregate multiscale contextual information and provide a large receptive field with a small number of trainable weights, which provide computational benefit. 2D convolutions are applied in the proposed model to enhance the computational efficiency. The constructed model is tested by performing a multi-class human thyroid tissue classification on 3D OCT images. For comparison, experimental results are obtained for texture feature-based shallow learning models and typical deep learning classification models. The results show that the proposed DDSCN model outperforms those state-of-art models and can improve accuracy by 3.2% compared with the best texture-based model and 2.27% compared with the best CNN model. The proposed deep model provides significant potential for the applicability of the deep learning technique to analyze medical images of human tissue while advancing the next generation of OCT-based real-time surgery image guidance.

Traffic Scene Depth Analysis Based on Depthwise Separable Convolutional Neural Network

In order to obtain the distances between the surrounding objects and the vehicle in the traffic scene in front of the vehicle, a monocular visual depth estimation method based on the depthwise separable convolutional neural network is proposed in this study. First, features containing shallow depth information were extracted from the RGB images using the convolution layers and maximum pooling layers. Subsampling operations were also performed on these images. Subsequently, features containing advanced depth information were extracted using a block based on an ensemble of convolution layers and a block based on depth separable convolution layers. The output from all different blocks is combined afterwards. Finally, transposed convolution layers were used for upsampling the feature maps to the same size with the original RGB image. During the upsampling process, skip connections were used to merge the features containing shallow depth information that was obtained from the convolution operation through the depthwise separable convolution layers. The depthwise separable convolution layers can provide more accurate depth information features for estimating the monocular visual depth. At the same time, they require reduced computational cost and fewer parameter numbers while providing a similar level (or slightly better) computing performance. Integrating multiple simple convolutions into a block not only increases the overall depth of the neural network but also enables a more accurate extraction of the advanced features in the neural network. Combining the output from multiple blocks can prevent the loss of features containing important depth information. The testing results show that the depthwise separable convolutional neural network provides a superior performance than the other monocular visual depth estimation methods. Therefore, applying depthwise separable convolution layers in the neural network is a more effective and accurate approach for estimating the visual depth.

A New Deep Learning Algorithm for SAR Scene Classification Based on Spatial Statistical Modeling and Features Re-Calibration.

Synthetic Aperture Radar (SAR) scene classification is challenging but widely applied, in which deep learning can play a pivotal role because of its hierarchical feature learning ability. In the paper, we propose a new scene classification framework, named Feature Recalibration Network with Multi-scale Spatial Features (FRN-MSF), to achieve high accuracy in SAR-based scene classification. First, a Multi-Scale Omnidirectional Gaussian Derivative Filter (MSOGDF) is constructed. Then, Multi-scale Spatial Features (MSF) of SAR scenes are generated by weighting MSOGDF, a Gray Level Gradient Co-occurrence Matrix (GLGCM) and Gabor transformation. These features were processed by the Feature Recalibration Network (FRN) to learn high-level features. In the network, the Depthwise Separable Convolution (DSC), Squeeze-and-Excitation (SE) Block and Convolution Neural Network (CNN) are integrated. Finally, these learned features will be classified by the Softmax function. Eleven types of SAR scenes obtained from four systems combining different bands and resolutions were trained and tested, and a mean accuracy of 98.18% was obtained. To validate the generality of FRN-MSF, five types of SAR scenes sampled from two additional large-scale Gaofen-3 and TerraSAR-X images were evaluated for classification. The mean accuracy of the five types reached 94.56%; while the mean accuracy for the same five types of the former tested 11 types of scene was 96%. The high accuracy indicates that the FRN-MSF is promising for SAR scene classification without losing generality.

3D Medical Image Classification with Depthwise Separable Networks

This research studies a dilated depthwise separable convolution neural network (DSCN) model to identify human tissue types from 3D medical images. 3D medical image classification is a challenging task due to the unpredictable noise and indistinct tissue behaviors of the image content. The objective of this research is to improve typical supervised deep learning model accuracy by using dilated convolution and depthwise separable network approaches on 3D medical image classification tasks. A depthwise separable architecture is used to improve parameter utilization efficiency. Dilated convolutions are applied to systematically aggregate multiscale contextual information and provide a large receptive field with a small number of trainable weights. The performance of the constructed model is tested to perform a multi-class human tissue classification on 3D Optical Coherence Tomography (OCT) images. Experimental results are compared with typical deep learning classification models. The results show that the DSCN model outperforms other models for all the tissue classification tasks. The proposed DSCN model can be a potential approach for the 3D image-based diagnostic tasks in both healthcare and manufacturing field.

Designing efficient accelerator of depthwise separable convolutional neural network on FPGA

In recent years, convolutional neural networks (CNNs) have achieved state-of-the-art results for many computer vision tasks. However, the traditional CNNs are computational-intensive and memory-intensive, hence they are unsuitable for the application in mobile edge computing scenarios with limited computing resources and low power consumption. The depthwise separable CNNs can significantly reduce the number of model parameters and improve the calculation speed, so it is naturally suitable for mobile edge computing applications. In this paper, we propose a Field Programmable Gate Array (FPGA)-based depthwise separable CNN accelerator with all the layers working concurrently in a pipelined fashion to improve the system throughput and performance. To implement the accelerator, we present a custom computing engine architecture to handle the dataflow between adjacent layers by using double-buffering-based memory channels. Besides, in fully connected layers, data titling technique is adopted to divide matrix multiplication from large dimension into small matrix. Finally, our proposed accelerator for depthwise separable CNN has been implemented and evaluated on Intel Arria 10 FPGA. The results of experiment indicate that the proposed depthwise separable CNN accelerator has a performance of 98.9 GOP/s and achieve up to 17.6× speed up and 29.4× low power than CPU and GPU implementations respectively.