Two-stage Convolutional Neural Network Research Articles

Automated Prediction of Malignant Melanoma using Two-Stage Convolutional Neural Network.

A skin lesion refers to an area of the skin that exhibits anomalous growth or distinctive visual characteristics compared to the surrounding skin. Benign skin lesions are noncancerous and generally pose no threat. These irregular skin growths can vary in appearance. On the other hand, malignant skin lesions correspond to skin cancer, which happens to be the most prevalent form of cancer in the United States. Skin cancer involves the unusual proliferation of skin cells anywhere on the body. The conventional method for detecting skin cancer is relatively more painful. This work involves the automated prediction of skin cancer and its types using two stage Convolutional Neural Network (CNN). The first stage of CNN extracts low level features and second stage extracts high level features. Feature selection is done using these two CNN and ABCD (Asymmetry, Border irregularity, Colour variation, and Diameter) technique. The features extracted from the two CNNs are fused with ABCD features and fed into classifiers for the final prediction. The classifiers employed in this work include ensemble learning methods such as gradient boosting and XG boost, as well as machine learning classifiers like decision trees and logistic regression. This methodology is evaluated using the International Skin Imaging Collaboration (ISIC) 2018 and 2019 dataset. As a result, the first stage CNN which is used for creation of new dataset achieved an accuracy of 97.92%. Second stage CNN which is used for feature selection achieved an accuracy of 98.86%. Classification results are obtained for both with and without fusion of features. Therefore, two stage prediction model achieved better results with feature fusion.

Convolutional neural network with parallel convolution scale attention module and ResCBAM for breast histology image classification

Breast cancer is the most common cause of female morbidity and death worldwide. Compared with other cancers, early detection of breast cancer is more helpful to improve the prognosis of patients. In order to achieve early diagnosis and treatment, clinical treatment requires rapid and accurate diagnosis. Therefore, the development of an automatic detection system for breast cancer suitable for patient imaging is of great significance for assisting clinical treatment. Accurate classification of pathological images plays a key role in computer-aided medical diagnosis and prognosis. However, in the automatic recognition and classification methods of breast cancer pathological images, the scale information, the loss of image information caused by insufficient feature fusion, and the enormous structure of the model may lead to inaccurate or inefficient classification. To minimize the impact, we proposed a lightweight PCSAM-ResCBAM model based on two-stage convolutional neural network. The model included a Parallel Convolution Scale Attention Module network (PCSAM-Net) and a Residual Convolutional Block Attention Module network (ResCBAM-Net). The first-level convolutional network was built through a 4-layer PCSAM module to achieve prediction and classification of patches extracted from images. To optimize the network's ability to represent global features of images, we proposed a tiled feature fusion method to fuse patch features from the same image, and proposed a residual convolutional attention module. Based on the above, the second-level convolutional network was constructed to achieve predictive classification of images. We evaluated the performance of our proposed model on the ICIAR2018 dataset and the BreakHis dataset, respectively. Furthermore, through model ablation studies, we found that scale attention and dilated convolution play an important role in improving model performance. Our proposed model outperforms the existing state-of-the-art models on 200 × and 400 × magnification datasets with a maximum accuracy of 98.74 %.

Two-Stage Convolutional Neural Network for Classification of Movement Patterns in Tremor Patients

Two-Stage Convolutional Neural Network for Classification of Movement Patterns in Tremor Patients

A Two-Stage Convolutional Neural Network for Interactive Channel Segmentation From 3-D Seismic Data

A Two-Stage Convolutional Neural Network for Interactive Channel Segmentation From 3-D Seismic Data

Assessment of approach separation with probabilistic aircraft wake vortex recognition via deep learning

Compared to static aircraft wake separation during landing/departure, reducing separation minima related to wake turbulence without compromising safety has initially demonstrated operational benefits, while dynamic pairwise separation remains under development. This paper proposes a two-stage probabilistic deep learning framework for wake vortex recognition and duration assessment, using wake images from the Light Detection and Ranging (LiDAR) instrument at Hong Kong International Airport. The first stage consists of vortex core locating utilising the Convolutional Neural Network (CNN), and the second stage predicts vortex strength within the Region of Interest (ROI), derived from raw images based on the initial core locating results. The existence of vortices is assessed upon the reliable probabilistic estimation of vortex movement under specific wind conditions and the estimation of its endurance in the final approach path. Furthermore, the prediction uncertainty is explained from the feature analysis aspect. Computational results indicate that the proposed two-stage CNN framework excels in estimating the spatial features and strength of coupled vortices. The wake duration measurement suggests a high potential for separation minima reduction when the crosswind exceeds (2–3)m/s. This will establishe conditions for onboard real-time wake monitoring, and the development of dynamic pairwise and meteorologically-related aircraft separation systems.

A Deep Learning Application for Building Damage Assessment Using Ultra-High-Resolution Remote Sensing Imagery in Turkey Earthquake

Rapid building damage assessment following an earthquake is important for humanitarian relief and disaster emergency responses. In February 2023, two magnitude-7.8 earthquakes struck Turkey in quick succession, impacting over 30 major cities across nearly 300 km. A quick and comprehensive understanding of the distribution of building damage is essential for efficiently deploying rescue forces during critical rescue periods. This article presents the training of a two-stage convolutional neural network called BDANet that integrated image features captured before and after the disaster to evaluate the extent of building damage in Islahiye. Based on high-resolution remote sensing data from WorldView2, BDANet used pre-disaster imagery to extract building outlines; the image features before and after the disaster were then combined to conduct building damage assessment. We optimized these results to improve the accuracy of building edges and analyzed the damage to each building, and used population distribution information to estimate the population count and urgency of rescue at different disaster levels. The results indicate that the building area in the Islahiye region was 156.92 ha, with an affected area of 26.60 ha. Severely damaged buildings accounted for 15.67% of the total building area in the affected areas. WorldPop population distribution data indicated approximately 253, 297, and 1,246 people in the collapsed, severely damaged, and lightly damaged areas, respectively. Accuracy verification showed that the BDANet model exhibited good performance in handling high-resolution images and can be used to directly assess building damage and provide rapid information for rescue operations in future disasters using model weights.

Automated deep learning-based AMD detection and staging in real-world OCT datasets (PINNACLE study report 5)

Real-world retinal optical coherence tomography (OCT) scans are available in abundance in primary and secondary eye care centres. They contain a wealth of information to be analyzed in retrospective studies. The associated electronic health records alone are often not enough to generate a high-quality dataset for clinical, statistical, and machine learning analysis. We have developed a deep learning-based age-related macular degeneration (AMD) stage classifier, to efficiently identify the first onset of early/intermediate (iAMD), atrophic (GA), and neovascular (nAMD) stage of AMD in retrospective data. We trained a two-stage convolutional neural network to classify macula-centered 3D volumes from Topcon OCT images into 4 classes: Normal, iAMD, GA and nAMD. In the first stage, a 2D ResNet50 is trained to identify the disease categories on the individual OCT B-scans while in the second stage, four smaller models (ResNets) use the concatenated B-scan-wise output from the first stage to classify the entire OCT volume. Classification uncertainty estimates are generated with Monte-Carlo dropout at inference time. The model was trained on a real-world OCT dataset, 3765 scans of 1849 eyes, and extensively evaluated, where it reached an average ROC-AUC of 0.94 in a real-world test set.

Empowering facial emotion recognition in service industry – a two-stage convolutional neural network model

Empowering facial emotion recognition in service industry – a two-stage convolutional neural network model

Two-stage coarse-to-fine image anomaly segmentation and detection model

Existing Convolutional Neural Network (CNN) based anomaly detection and segmentation approaches are overly sensitive or not sensitive enough to noise, resulting in anomaly patterns, partially detected in the testing stage. The previous methods may also differentiate normal and abnormal images, but they cannot identify the location of anomaly presented in test images with high accuracy. To address this issue, we propose a two-stage CNN model for coarse-to-fine anomaly segmentation and detection called (TASAD). In both stages of TASAD, we train our model on a mixture of normal and abnormal training samples. The abnormal images are obtained by inserting pseudo-anomaly patterns that are automatically generated from anomaly source images. We use a novel and sophisticated anomaly insertion technique to generate various anomalous samples. In the first stage, we design a coarse anomaly segmentation (CAS) model that takes a whole image as an input, while in the second stage, we train a fine anomaly segmentation (FAS) model on image patches. FAS model improves detection and segmentation performance by refining anomaly patterns partially detected by CAS model. We train our framework on MVTec dataset and compare it with state-of-the-art (SOTA) methods. The proposed architecture leads to a compact model size – four times smaller than the SOTA method, while exhibiting better pixel-level accuracy. TASAD can also be applied to SOTAs to further improve their anomaly detection performance. Our experiments demonstrate that when applied to the latest SOTAs, TASAD improves the average precision (AP) performance of previous methods by 6.2%. For reproducibility of the results, code is provided at https://github.com/RizwanAliQau/tasad.git.

Joint two-stage convolutional neural networks for intracranial aneurysms detection on 3D TOF-MRA

Objective. This study aims to develop a three-dimensional convolutional neural network utilizing computer-aided diagnostic technology to facilitate the detection of intracranial aneurysms and automatically assess their location and extent, thereby enhancing the efficiency of radiologists, and streamlining clinical workflows. Approach. A retrospective study was conducted, proposing a joint segmentation and classification network (JSCD-Net) that employs 3D time-of-flight magnetic resonance angiography images for preliminary detection of aneurysms and the minimization of false positives. Specifically, the U-Net++ network was utilized for pre-detection of aneurysms. This was followed by the creation of a multi-path network, co-trained with U-Net++ to correct the results of the first stage to further reduce the rate of false positives. Model effectiveness and robustness were evaluated using sensitivity and false positive analyses on internal and external datasets. A cross-validated free-response receiver operating characteristic curve was also plotted. Main results. JSCD-Net demonstrated a sensitivity of 91.2% (31 of 34; 95% CI: 77.0, 97.0) with an average of 3.55 false positives per scan on the internal test set. For the external test set, it identified 97.2% (70 of 72; 95% CI: 90.4, 99.2) of aneurysms with an average of 2.7 false positives per scan. Significance. When compared with the existing studies, the proposed model shows high sensitivity in detecting intracranial aneurysms with a reasonable number of false positives per case. This result emphasizes the model’s potential as a valuable tool in aiding clinical diagnoses.

Vehicle Detection in Multisource Remote Sensing Images Based on Edge-Preserving Super-Resolution Reconstruction

As an essential technology for intelligent transportation management and traffic risk prevention and control, vehicle detection plays a significant role in the comprehensive evaluation of the intelligent transportation system. However, limited by the small size of vehicles in satellite remote sensing images and lack of sufficient texture features, its detection performance is far from satisfactory. In view of the unclear edge structure of small objects in the super-resolution (SR) reconstruction process, deep convolutional neural networks are no longer effective in extracting small-scale feature information. Therefore, a vehicle detection network based on remote sensing images (VDNET-RSI) is constructed in this article. The VDNET-RSI contains a two-stage convolutional neural network for vehicle detection. In the first stage, a partial convolution-based padding adopts the improved Local Implicit Image Function (LIIF) to reconstruct high-resolution remote sensing images. Then, the network associated with the results from the first stage is used in the second stage for vehicle detection. In the second stage, the super-resolution module, detection heads module and convolutional block attention module adopt the increased object detection framework to improve the performance of small object detection in large-scale remote sensing images. The publicly available DIOR dataset is selected as the experimental dataset to compare the performance of VDNET-RSI with that of the state-of-the-art models in vehicle detection based on satellite remote sensing images. The experimental results demonstrated that the overall precision of VDNET-RSI reached 62.9%, about 6.3%, 38.6%, 39.8% higher than that of YOLOv5, Faster-RCNN and FCOS, respectively. The conclusions of this paper can provide a theoretical basis and key technical support for the development of intelligent transportation.

Robust Adaptive Beamforming Based on a Convolutional Neural Network

To address the advancements in jamming technology, it is imperative to consider robust adaptive beamforming (RBF) methods with finite snapshots and gain/phase (G/P) errors. This paper introduces an end-to-end RBF approach that utilizes a two-stage convolutional neural network. The first stage includes convolutional blocks and residual blocks without downsampling; the blocks assess the covariance matrix precisely using finite snapshots. The second stage maps the first stage’s output to an adaptive weight vector employing a similar structure to the first stage. The two stages are pre-trained with different datasets and fine-tuned as end-to-end networks, simplifying the network training process. The two-stage structure enables the network to possess practical physical meaning, allowing for satisfying performance even with a few snapshots in the presence of array G/P errors. We demonstrate the resulting beamformer’s performance with numerical examples and compare it to various other adaptive beamformers.

Weld Defect Monitoring Based on Two-Stage Convolutional Neural Network

Weld Defect Monitoring Based on Two-Stage Convolutional Neural Network

Classification and regression-based convolutional neural network and long short-term memory configuration for bridge damage identification using long-term monitoring vibration data

Considerable attention has recently been focused on classification and regression-based convolutional neural network (CNN) and long short-term memory (LSTM) due to their excellent performance in capturing complex spatial and temporal information characteristics for structural damage identification. However, few studies have considered structural damage identification as a classification and regression problem. In addition, bridges in practical engineering are vulnerable to various environmental and vehicle loading conditions. Hence, this study proposed a new two-stage CNN–LSTM configuration for bridge damage identification using vibration data considering the influence of temperatures. First, a classification-based CNN–LSTM is designed to perform multiclass damage detection tasks, and then a regression-based CNN–LSTM is developed for damage localization and severity prediction tasks. The performance of the proposed damage identification method was evaluated through a simulation dataset of a concrete highway bridge model and a field experiment dataset of Z24-bridge (Switzerland). In addition, a set of statistical evaluation metrics such as sparse categorical cross-entropy loss, accuracy, confusion matrix, mean squared loss, mean absolute error, mean absolute percentage error, and coefficient of determination were used to compare the damage identification performance of the proposed CNN–LSTM configuration with a regular CNN model and conventional machine learning (ML) algorithms. Prediction results indicate that the proposed CNN–LSTM model outperforms the regular CNN model and conventional ML algorithms for bridge damage identification.

Two-stage convolutional neural network for joint removal of sensor noise and background interference in lock-in thermography

Lock-in thermography provides a fast, low-cost, and accurate non-destructive testing for subsurface defect detection in materials. However, there is serious sensor noise and background interference presented in the infrared images captured by lock-in thermography. In this paper, we set up lock-in thermography experiments to analyze the characteristics of sensor noise and background interference. Based on their different characteristics, we design a two-stage CNN model for the joint removal of sensor noise and background interference. This model consists of a number of multi-direction feature extraction blocks (MDFEBs) to extract local features in different directions for pixel-level sensor noise removal, and an encoder–decoder architecture with large receptive fields to extract global features for background interference correction. Experimental results demonstrate that our proposed method outperforms the well-established infrared image denoising and non-uniform background correction techniques on specimens made of different materials and at various excitation frequencies. The proposed two-stage CNN model provides high-quality input of infrared images for the following subsurface defect detection task in lock-in thermography.

Using image analysis to quantify defects and prioritize repairs in built-up roofs

PurposeThis paper aims to use convolutional neural networks (CNNs) to provide an objective approach to classify deteriorated building assets according to the type and extent of damage. This research supports automated inspection of buildings and focuses on roofing elements as one of the most critical and externally distressed elements in buildings.Design/methodology/approachIn this paper, 5,000+ images of deteriorated roofs from several buildings were collected to design a CNN system that automatically identifies and sizes roofing defects. Experimenting with different CNN formulations, the best accuracy is achieved using two-stage CNNs. The first-stage CNN classifies images into defect/no defect, while the second stage classifies the defected images according to the damage type. Based on the image classification, optimization is used to prioritize roof repairs by maximizing the return from limited rehabilitation funds.FindingsThe developed CNNs reached 95% and 97% accuracy for the first and second phases, respectively, which is higher than achieved in previous literature efforts. Using the proposed model to automate inspection and condition assessment activities proved to be faster than conventional methods. Repair/replace strategy for a case study of 21 campus buildings based on their condition and budgetary constraints was suggested.Research limitations/implicationsFuture research includes testing different data acquisition technologies (e.g. infrared imaging), performing severity-based classification and integrating with BIM for defect localization.Originality/valueThis study provides an objective approach to automate asset condition assessment and improve funding decisions using a combination of image analysis and optimization techniques. The proposed approach is applicable toward other asset types and components.

Hierarchical Parallel Vision Processor for High-Speed Ship Detection

Ship detection is essential in ship rescue and marine traffic safety. However, high-speed real-time ship detection has become problematic in practical applications. In this brief, we propose a high-performance ship detection system utilizing the co-design of algorithm and hardware. First, the proposed Coarse-to-Fine Classification and Segmentation algorithm contains a two-stage convolutional neural network. It quickly locates the Region of Interest (RoI) and then accurately locates the ship’s position in the RoI without low-parallelism non-maximum suppression (NMS) post-processing. Second, we proposed a Hierarchical Parallel Vision Processor, including a pixel-parallel processing unit, a patch-parallel processing unit, and a global micro-processing unit, ensuring high-speed processing with different parallelism. Finally, we designed a high-speed real-time ship detection system. The experimental results show that our ship detection system based on the frequency of 100Mhz on Arria 10 FPGA can achieve a speed of 4778FPS, which is beneficial in practical applications.

EYOLOv3: An Efficient Real-Time Detection Model for Floating Object on River

At present, the surveillance of river floating in China is labor-intensive, time-consuming, and may miss something, so a fast and accurate automatic detection method is necessary. The two-stage convolutional neural network models appear to have high detection accuracy, but it is hard to reach real-time detection, while on the other hand, the one-stage models are less time-consuming but have lower accuracy. In response to the above problems, we propose a one-stage object detection model EYOLOv3 to achieve real-time and high accuracy detection of floating objects in video streams. Firstly, we design a multi-scale feature extraction and fusion module to improve the feature extraction capability of the network. Secondly, a better clustering algorithm is used to analyze the size characteristics of floating objects to design the anchor box, enabling the network to detect objects more effectively. Then a focus loss function is proposed to make the network effectively overcome the sample imbalance problem, and finally, an improved NMS algorithm is proposed to solve the object suppressed problem. Experiments show that the proposed model is efficient in detection of river floating objects, and has better performance than the classical object detection method and the latest method, realizing real-time floating detection in video streams.

High-precision automatic identification method for dicentric chromosome images using two-stage convolutional neural network

Dicentric chromosome analysis is the gold standard for biological dose assessment. To enhance the efficiency of biological dose assessment in large-scale radiation catastrophes, automatic identification of dicentric chromosome images is a promising and objective method. In this paper, an automatic identification method for dicentric chromosome images using two-stage convolutional neural network is proposed based on Giemsa-stained automatic microscopic imaging. To automatically segment the adhesive chromosome masses, a k-means based adaptive image segmentation and watershed segmentation algorithm is applied. The first-stage CNN is used to identify the dicentric chromosome images from all the images and the second-stage CNN works to specifically identify the dicentric chromosome images. This two-stage CNN identification method can effectively detects chromosome images with concealed centromeres, poorly expanded and long-armed entangled chromosomes, and tricentric chromosomes. The novel two-stage CNN method has a chromosome identification accuracy of 99.4%, a sensitivity of 85.8% sensitivity, and a specificity of 99.6%, effectively reducing the false positive rate of dicentric chromosome. The analysis speed of this automatic identification method can be 20 times quicker than manual detection, providing a valuable reference for other image identification situations with small target rates.

Detection of Periapical Lesions on Panoramic Radiographs Using Deep Learning

Dentists could fail to notice periapical lesions (PLs) while examining panoramic radiographs. Accordingly, this study aimed to develop an artificial intelligence (AI) designed to address this problem. Materials and methods: a total of 18618 periapical root areas (PRA) on 713 panoramic radiographs were annotated and classified as having or not having PLs. An AI model consisting of two convolutional neural networks (CNNs), a detector and a classifier, was trained on the images. The detector localized PRAs using a bounding-box-based object detection model, while the classifier classified the extracted PRAs as PL or not-PL using a fine-tuned CNN. The classifier was trained and validated on a balanced subset of the original dataset that included 3249 PRAs, and tested on 707 PRAs. Results: the detector achieved an average precision of 74.95%, while the classifier accuracy, sensitivity and specificity were 84%, 81% and 86%, respectively. When integrating both detection and classification models, the proposed method accuracy, sensitivity, and specificity were 84.6%, 72.2%, and 85.6%, respectively. Conclusion: a two-stage CNN model consisting of a detector and a classifier can successfully detect periapical lesions on panoramic radiographs.