Fig 1 - available via license: Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Content may be subject to copyright.
Source publication
Its crucial for financial systems to have sound security measures in place. For security reasons customers are not allowed to wear a helmet while using ATM(Automated Teller Machine). An automated helmet detection using ATM surveillance camera feed can help improve security significantly. Recently deep convolutional neural network (DCNN) have shown...
Contexts in source publication
Context 1
... of a larger, more abstract and perhaps more meaningful discrete object category (such as a man with a Helmet, a man without a Helmet), rather than just the sum of their parts. Furthermore, a judgment is made as to whether those objects are worthy of our focus and attention in the overall image presented before us; the man in this picture (see Fig. 1) clearly the object of interest in this picture is the user, whereas the cash machine and the poster are part of the background. It is often desirable for digital images to be treated by the computer in a similar manner. The process by which we try to classify a digital image into different categories is of particular interest to many ...
Context 2
... each model is pretrained on ImageNet and transfer learning are used on all of them over the same proprietary ATM dataset under similar conditions. At each epoch training accuracy and testing accuracy of these models are checked with a randomly picked labeled image that doesn't include in the current training or testing batch respectively. (see Fig. 10). Table 2 shows the experimental results about the helmet recognition accuracy of transfer learned inception model with comparison to VGG-16 [16] and Resnet 152 [5]. Training deep neural networks on smaller dataset was challenging but it turns out to give better results in detection because of the huge number of features readily ...
Similar publications
The most common type of transportation in India is two-wheeler, where the number of accidents are increasing day-by-day. In general, these accidents have occurred due to riding a motorcycle without a helmet. It is very difficult to monitor each and every rider whether they are wearing a helmet or not, by a human labor, where as an electronic detect...
Citations
... Identifying heritage sites, detecting cracks in concrete [64], and recognizing scholars and well-known individuals from images are significant tasks in image classification. In a research study [65], the authors employed transfer learning to incorporate a preexisting CNN model for heritage data classification. Fabric defect identification can also be addressed using computer vision (CV). ...
The computer vision industry has experienced a significant surge in growth, resulting in numerous promising breakthroughs in computer intelligence. The present review paper outlines the advantages and potential future implications of utilizing this technology in education. A total of 84 research publications have been thoroughly scrutinized and analyzed. The study revealed that computer vision technology integrated with a mobile application is exceptionally useful in monitoring students’ perceptions and mitigating academic dishonesty. Additionally, it facilitates the digitization of handwritten scripts for plagiarism detection and automates attendance tracking to optimize valuable classroom time. Furthermore, several potential applications of computer vision technology for educational institutions have been proposed to enhance students’ learning processes in various faculties, such as engineering, medical science, and others. Moreover, the technology can also aid in creating a safer campus environment by automatically detecting abnormal activities such as ragging, bullying, and harassment.
... Transfer learning can help reduce the amount of training data needed to train efficient deep learning models. Transfer learning [12] allows the transfer of what state-of-the-art machine learning models have learned to custom problems. The state-of-the-art machine learning models are trained with an enormous amount of data and are generally good at recognizing objects in images. ...
Fast and automatic extraction of buildings from aerial imagery is essential for various applications. However, it remains a challenge, mainly because it requires the accurate recovery of buildings from high-resolution aerial imagery data. The task is particularly challenging due to the distinct nature of building shapes. The extraction of buildings from aerial imagery is often manually conducted by humans, which makes the process very time-consuming and costly. Furthermore, it is especially challenging to automatically extract the footprints of closely spaced buildings in dense urban areas through traditional segmentation techniques. This study proposes an instance segmentation-based building extraction technique for effective building detection and segmentation. The proposed method combines transfer learning with Mask Regional Convolutional Neural Network (Mask R-CNN) integrated with PointRend to locate and generate high-quality segmentation masks for building instances. In addition, several data augmentation strategies are also implemented to enlarge the training dataset. Results indicate that the proposed instance segmentation-based method for building extraction is able to execute distinctly efficient and effective automatic segmentation of distinctive and complex-shaped buildings in a dense urban area.
... Pang et al used a transfer learning approach and a deep learning model based on the basic picture elements of real medical images using supervised training [12]. Alwyn Mathew et al used pre-trained Google's inception model for ATM surveillance dataset for security purpose [13]. For an image search and classifications, The pre-trained CNNs, like VGG-Net and AlexNet are used to extract the features [14]. ...
: Medical imaging plays an important role in the diagnosis of some critical diseases and further treatment process of patients. Brain is a central and most complex structure in the human body that works with billions of cells, which controls all other organ functioning. Brain tumours observed as uncontrolled abnormal cell growth in brain tissues. Classification of such cells in a early stage will increase the survival rate of the patient. Machine learning algorithms have contributed much in automation of such tasks. Further improvement in prediction rate is possible through deep learning models. In this paper presents experiments by deep transfer learning models on publicly available dataset for Brain tumour classification. Pre-trained plain and residual feed forward models such as Alexnet, VGG19, ResNet50, ResNet101 and GoogleNet are used for the purpose of feature extraction, Fully connected layers and softmax layer for classification is used commonly. The evaluation metrics Accuracy, Sensitivity, Specificity and F1-Score were computed.
... Following Inception-V1, three variants have been introduced to refine the architecture by (i) using batch normalization for training in Inception-V2 (Ioffe and Szegedy, 2015), (ii) using a factorization approach to reduce the computational cost of convolution layers in Inception-V3 (Szegedy et al., 2016), and (iv) introducing a simplified uniform variant of Inception-v3 with more inception modules in Inception V-4 (Szegedy et al., 2017). For instance, in Mathew et al. (2017), the knowledge learned by Inception v3 on 1,28 million images (categorized into 1000 classes) from the ImageNet LSVRC 2014 is utilized for intrusion detection on small ATM surveillance video dataset of 4719 images. • ResNet: augmenting CNN models with more layers can result in vanishing gradients and accuracy saturation. ...
Recently, developing automated video surveillance systems (VSSs) has become crucial to ensure the security and safety of the population, especially during events involving large crowds, such as sporting events. While artificial intelligence (AI) smooths the path of computers to think like humans, machine learning (ML) and deep learning (DL) pave the way more, even by adding training and learning components. DL algorithms require data labeling and high-performance computers to effectively analyze and understand surveillance data recorded from fixed or mobile cameras installed in indoor or outdoor environments. However, they might not perform as expected, take much time in training, or not have enough input data to generalize well. To that end, deep transfer learning (DTL) and deep domain adaptation (DDA) have recently been proposed as promising solutions to alleviate these issues. Typically, they can (i) ease the training process, (ii) improve the generalizability of ML and DL models, and (iii) overcome data scarcity problems by transferring knowledge from one domain to another or from one task to another. Although the increasing number of articles proposed to develop DTL- and DDA-based VSSs, a thorough review that summarizes and criticizes the state-of-the-art is still missing. To that end, this paper introduces, to the best of the authors’ knowledge, the first overview of existing DTL- and DDA-based video surveillance to (i) shed light on their benefits, (ii) discuss their challenges, and (iii) highlight their future perspectives.
... Though a significant amount of training data is required for deep learning, finding enough data is difficult within a certain range for a specific problem. Therefore, transfer learning is proposed as a solution for this problem [28]. It performs new model training by transferring the source model parameters to new target model. ...
In this paper, a fast wind turbine defect detection model is proposed with a Cascade Mask region Convolutional Neural network (Cascade Mask R-CNN). Instead of standard convolution in the backbone network of Cascade Mask R-CNN, a depthwise separable convolution is used to minimize the computation cost. Moreover, image augmentation and transfer learning techniques are also involved to enhance the performance of the proposed model. The detection and instance segmentation performance of the proposed model is compared with existing techniques in terms of mean intersection over union (MIoU), mean average precision (MAP) and classifier accuracy. The experimental results show that the proposed WTB defect detection and classification model shows better performance with 82.42% MAP, 87.49% MIoU and 97.8% classifier accuracy.
... The use of ATM surveillance camera feed can help improve security significantly as a form of automated helmet detection. The model achieved an accuracy of 95.3% whilst training on a proprietary ATM surveillance dataset [37]. ...
... Important fields, such as medicine, industry and finance, are high-value objectives. Effective IDSs must be designed to efficiently and intelligently interpret intrusion attempts in incoming network traffic energy [8,18,37,43,50,74]. ...
... A big dataset is crucial and recommended by researchers to enhance detection accuracy [37]. An experiment was conducted by collecting packets from a small network dataset that can be expanded only once [19,34,71,72]. ...
This study reviews and analyses the research landscape for intrusion detection systems (IDSs) based on deep learning (DL) techniques into a coherent taxonomy and identifies the gap in this pivotal research area. The focus is on articles related to the keywords ‘deep learning’, ‘intrusion’ and ‘attack’ and their variations in four major databases, namely Web of Science, ScienceDirect, Scopus and the Institute of Electrical and Electronics Engineers’ Xplore. These databases are sufficiently broad to cover the technical literature. The dataset comprises 68 articles. The largest proportion (72.06%; 49/68) relates to articles that develop an approach for evaluating or identifying intrusion detection techniques using the DL approach. The second largest proportion (22.06%; 15/68) relates to studying/applying articles to the DL area, IDSs or other related issues. The third largest proportion (5.88%; 4/68) discusses frameworks/models for running or adopting IDSs. The basic characteristics of this emerging field are identified from the aspects of motivations, open challenges that impede the technology’s utility, authors’ recommendations and substantial analysis. Then, a result analysis mapping for new directions is discussed. Three phases are designed to meet the demands of detecting distributed denial-of-service attacks with a high accuracy rate. This study provides an extensive resource background for researchers who are interested in IDSs based on DL.
... Deep learning requires a significant amount of data, but in most cases it is difficult to find enough training data for a specific problem within a certain range. To solve this problem, a solution is proposed, namely to use transfer learning [24]. ...
Traffic congestion due to vehicular accidents seriously affects normal travel, and accurate and effective mitigating measures and methods must be studied. To resolve traffic accident compensation problems quickly, a vehicle-damage-detection segmentation algorithm based on transfer learning and an improved mask regional convolutional neural network (Mask RCNN) is proposed in this paper. The experiment first collects car damage pictures for preprocessing and uses Labelme to make data set labels, which are divided into training sets and test sets. The residual network (ResNet) is optimized, and feature extraction is performed in combination with Feature Pyramid Network (FPN). Then, the proportion and threshold of the Anchor in the region proposal network (RPN) are adjusted. The spatial information of the feature map is preserved by bilinear interpolation in ROIAlign, and different weights are introduced in the loss function for different-scale targets. Finally, the results of self-made dedicated dataset training and testing show that the improved Mask RCNN has better Average Precision (AP) value, detection accuracy and masking accuracy, and improves the efficiency of solving traffic accident compensation problems.
... Otherwise; [11]. The precision was 95.3%. ...
The process of identifying food items from an image is one of the promising applications of visual object recognition in computer vision. However, the analysis of food items is
particularly challenging work due to the characteristic of their images, which is why a low classification accuracy has been achieved by traditional approaches in the field. Deep neural networks have exceeded such solutions. With a goal to successfully applying computer images, which is why a low classification accuracy vision techniques to classify food images based on Inception-v3 model of TensorFlow platform, we use the transfer learning technology to retrain the food category datasets. Our approach shows auspicious results with an average accuracy of 95.2% approximately in correctly recognizing among 7 traditional Bangladeshi foods.
... Pang et al used a transfer learning approach and a deep learning model based on the basic picture elements of real medical images using supervised training [12]. Alwyn Mathew et al used pre-trained Google's inception model for ATM surveillance dataset for security purpose [13]. For an image search and classifications, The pre-trained CNNs, like VGG-Net and AlexNet are used to extract the features [14]. ...
Medical imaging plays an important role in the diagnosis of some critical diseases and further treatment process of patients. Brain is a central and most complex structure in the human body that works with billions of cells, which controls all other organ functioning. Brain tumours observed as uncontrolled abnormal cell growth in brain tissues. Classification of such cells in a early stage will increase the survival rate of the patient. Machine learning algorithms have contributed much in automation of such tasks. Further improvement in prediction rate is possible through deep learning models. In this paper presents experiments by deep transfer learning models on publicly available dataset for Brain tumour classification. Pre-trained plain and residual feed forward models such as Alexnet, VGG19, ResNet50, ResNet101 and GoogleNet are used for the purpose of feature extraction, Fully connected layers and softmax layer for classification is used commonly. The evaluation metrics Accuracy, Sensitivity, Specificity and F1-Score were computed.
... It was proposed in works [8,9] to perform fine-tuning of object detector based on convolutional network VGG-16 using mini-batch stochastic gradient descent. However, this would require a significant volume of training set and a few days of working on the graphics processing unit for successful training. ...
Context. Lightweight model and effective training algorithm of on-board object detection system for a compact drone are developed. The object of research is the process of small object detection on aerial images under computational resource constraint and uncertainty caused by small amount of labeled training data. The subject of the research are the model and training algorithm for detecting small objects on aerial imagery.Objective. Goal of the research is developing efficient model and training algorithm of object detection system for a compact aerial drone under conditions of restricted computing resources and the limited volume of the labeled training set.Methods. The four-stage training algorithm of the object detector is proposed. At the first stage, selecting the type of deep convolutional neural network and the number of low-level layers that is pre-trained on the ImageNet dataset for reusing takes place. The second stage involves unsupervised training of high-level convolutional sparse coding layers using modification of growing neural gas to automatically determine the required number of neurons and provide optimal distributions of the neurons over the data. Its application makes it possible to utilize the unlabeled training datasets for the adaptation of the high-level feature description to the domain application area. At the third stage, there are reduction of output feture map using principan component analysis and building of decision rules. In this case, output feature map is formed by concatenation of feature maps from different level of deep network using upscaling upper maps to uniform shape for each channel. This approach provides more contextual information for efficient recognition of small objects on aerial images. In order to build classifier of output feature map pixels is proposed to use boostedinformation-extreme learning. Besides that, the extreme learning machine is used to build of regression model for predict bounding box of detected object. Last stage involves fine-tuning of high-level layers of deep network using simulated annealing metaheuristic algorithm in order to approximating the global optimum of complex criterion of training efficiency of detection model.Results. The results obtained on open datasets testify to the suitability of the model and training algorithm for practical usage. The proposed training algorithm utilize 500 unlabeled and 200 labeled training samples to provide 96% correctly detection of objects on the images of the test dataset.Conclusions. Scientific novelty of the paper is a new model and training algorithm for object detection, which enable to achieve high confidence of recognition of small objects on aerial images under computational resource constraint and limited volume of the labeled training set. Practical significance of the paper results consists in the developed model and training algorithm made it possible to reduce requirements for size of labeled training set and computation resources of on-board detection system of aerial drone in training and inference modes.
