• Journal of IKEEE
• /
• v.23 no.1
• /
• pp.242-248
• /
• 2019

Fire flame and smoke detection algorithm studies are challenging task in computer vision due to the variety of shapes, rapid spread and colors. The performance of a typical sensor based fire detection system is largely limited by environmental factors (indoor and fire locations). To solve this problem, a deep learning method is applied. Because it extracts the feature of the object using several methods, so that if a similar shape exists in the frame, it can be detected as false postive. This study proposes a new algorithm to reduce false positives by using frame similarity before using deep learning to decrease the false detection rate. Experimental results show that the fire detection performance is maintained and the false positives are reduced by applying the proposed method. It is confirmed that the proposed method has excellent false detection performance.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.104-107
• /
• 2019

Modern people have abundant food and various forms of life compared to the past, but they have come to form an unhealthy diet, such as skipping breakfast and not eating in time in a busy life. When these eating habits are maintained for a long time, it leads to digestive trouble. The most easily occurring symptoms are called reflux esophagitis and dysphagia. Among them, dysphagia requires quick and accurate diagnosis as they develop into various forms of complications or are also identified as presymptoms of gastric and laryngeal cancers. The result of the diagnosis is still passively judged by the doctor and each of results are different depending on the doctor. The result of the diagnosis here means the severity. When they identify treatment or complications following the results of the diagnosis, the wrong diagnosis may lead to excessive or insufficient treatment. In this paper, to figure out the severity of dysphagia in the diagnosis of dysphagia, we studied the development of a program using the K-means algorithm in the processing of X-ray images for identifying residual food in epiglottic vallecula and pyriform sinus in the section leading to esophagus.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.751-770
• /
• 2019

Action recognition is an essential task in computer vision due to the variety of prospective applications, such as security surveillance, machine learning, and human-computer interaction. The availability of more video data than ever before and the lofty performance of deep convolutional neural networks also make it essential for action recognition in video. Unfortunately, limited crafted video features and the scarcity of benchmark datasets make it challenging to address the multi-person action recognition task in video data. In this work, we propose a deep convolutional neural network-based Effective Hybrid Learning (EHL) framework for two-person interaction classification in video data. Our approach exploits a pre-trained network model (the VGG16 from the University of Oxford Visual Geometry Group) and extends the Faster R-CNN (region-based convolutional neural network a state-of-the-art detector for image classification). We broaden a semi-supervised learning method combined with an active learning method to improve overall performance. Numerous types of two-person interactions exist in the real world, which makes this a challenging task. In our experiment, we consider a limited number of actions, such as hugging, fighting, linking arms, talking, and kidnapping in two environment such simple and complex. We show that our trained model with an active semi-supervised learning architecture gradually improves the performance. In a simple environment using an Intelligent Technology Laboratory (ITLab) dataset from Inha University, performance increased to 95.6% accuracy, and in a complex environment, performance reached 81% accuracy. Our method reduces data-labeling time, compared to supervised learning methods, for the ITLab dataset. We also conduct extensive experiment on Human Action Recognition benchmarks such as UT-Interaction dataset, HMDB51 dataset and obtain better performance than state-of-the-art approaches.

• The Journal of the Korea Contents Association
• /
• v.21 no.8
• /
• pp.1-9
• /
• 2021

Deep learning in computer vision has made accelerated improvement over a short period but large-scale learning data and computing power are still essential that required time-consuming trial and error tasks are involved to derive an optimal network model. In this study, we propose a similar image classification performance improvement method based on CR (Confusion Rate) that considers only the characteristics of the data itself regardless of network optimization or data reinforcement. The proposed method is a technique that improves the performance of the deep learning model by calculating the CRs for images in a dataset with similar characteristics and reflecting it in the weight of the Loss Function. Also, the CR-based recognition method is advantageous for image identification with high similarity because it enables image recognition in consideration of similarity between classes. As a result of applying the proposed method to the Resnet18 model, it showed a performance improvement of 0.22% in HanDB and 3.38% in Animal-10N. The proposed method is expected to be the basis for artificial intelligence research using noisy labeled data accompanying large-scale learning data.

• Journal of the Earthquake Engineering Society of Korea
• /
• v.25 no.2
• /
• pp.71-81
• /
• 2021

This paper presents a real-time, false-pick filter based on deep learning to reduce false alarms of an onsite Earthquake Early Warning (EEW) system. Most onsite EEW systems use P-wave to predict S-wave. Therefore, it is essential to properly distinguish P-waves from noises or other seismic phases to avoid false alarms. To reduce false-picks causing false alarms, this study made the EEWNet Part 1 'False-Pick Filter' model based on Convolutional Neural Network (CNN). Specifically, it modified the Pick_FP (Lomax et al.) to generate input data such as the amplitude, velocity, and displacement of three components from 2 seconds ahead and 2 seconds after the P-wave arrival following one-second time steps. This model extracts log-mel power spectrum features from this input data, then classifies P-waves and others using these features. The dataset consisted of 3,189,583 samples: 81,394 samples from event data (727 events in the Korean Peninsula, 103 teleseismic events, and 1,734 events in Taiwan) and 3,108,189 samples from continuous data (recorded by seismic stations in South Korea for 27 months from 2018 to 2020). This model was trained with 1,826,357 samples through balancing, then tested on continuous data samples of the year 2019, filtering more than 99% of strong false-picks that could trigger false alarms. This model was developed as a module for USGS Earthworm and is written in C language to operate with minimal computing resources.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.12
• /
• pp.3904-3922
• /
• 2022

As a research hotspot, pedestrian detection has a wide range of applications in the field of computer vision in recent years. However, current pedestrian detection methods have problems such as insufficient detection accuracy and large models that are not suitable for large-scale deployment. In view of these problems mentioned above, a lightweight pedestrian detection and early warning method using a new model called you only look once (Yolov5) is proposed in this paper, which utilizing advantages of Yolov5s model to achieve accurate and fast pedestrian recognition. In addition, this paper also optimizes the loss function of the batch normalization (BN) layer. After sparsification, pruning and fine-tuning, got a lot of optimization, the size of the model on the edge of the computing power is lower equipment can be deployed. Finally, from the experimental data presented in this paper, under the training of the road pedestrian dataset that we collected and processed independently, the Yolov5s model has certain advantages in terms of precision and other indicators compared with traditional single shot multiBox detector (SSD) model and fast region-convolutional neural network (Fast R-CNN) model. After pruning and lightweight, the size of training model is greatly reduced without a significant reduction in accuracy, and the final precision reaches 87%, while the model size is reduced to 7,723 KB.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.3
• /
• pp.701-720
• /
• 2023

Aerial objects are more challenging to segment than normal objects, which are usually smaller and have less textural detail. In the process of segmentation, target objects are easily omitted and misdetected, which is problematic. To alleviate these issues, we propose local aggregation feature pyramid networks (LAFPNs) and pyramid integrated context modules (PICMs) for aerial object segmentation. First, using an LAFPN, while strengthening the deep features, the extent to which low-level features interfere with high-level features is reduced, and numerous dense and small aerial targets are prevented from being mistakenly detected as a whole. Second, the PICM uses global information to guide local features, which enhances the network's comprehensive understanding of an entire image and reduces the missed detection of small aerial objects due to insufficient texture information. We evaluate our network with the MS COCO dataset using three categories: airplanes, birds, and kites. Compared with Mask R-CNN, our network achieves performance improvements of 1.7%, 4.9%, and 7.7% in terms of the AP metrics for the three categories. Without pretraining or any postprocessing, the segmentation performance of our network for aerial objects is superior to that of several recent methods based on classic algorithms.

• Convergence Security Journal
• /
• v.22 no.5
• /
• pp.3-9
• /
• 2022

Recently, in order to build a cyber physical system(CPS) that is a technology related to the 4th industry, the construction of the virtual control system for physical model and control circuit simulation is increasingly required in various industries. It takes a lot of time and money to convert documents that are not electronically documented through direct input. For this, it is very important to digitize a large number of drawings that have already been printed through object recognition using artificial intelligence. In this paper, in order to accurately recognize objects in drawings and to utilize them in various applications, a recognition technique using artificial intelligence by analyzing the characteristics of objects in drawing was proposed. In order to improve the performance of object recognition, each object was recognized and then an intermediate file storing the information was created. And the recognition rate of the next recognition target was improved by deleting the recognition result from the drawing. In addition, the recognition result was stored as a standardized format document so that it could be utilized in various fields of the control system. The excellent performance of the technique proposed in this paper was confirmed through the experiments.

• International conference on construction engineering and project management
• /
• 2022.06a
• /
• pp.877-885
• /
• 2022

Vision-based activity recognition has been widely attempted at construction sites to estimate productivity and enhance workers' health and safety. Previous studies have focused on extracting an individual worker's postural information from sequential image frames for activity recognition. However, various trades of workers perform different tasks with similar postural patterns, which degrades the performance of activity recognition based on postural information. To this end, this research exploited a concept of human-object interaction, the interaction between a worker and their surrounding objects, considering the fact that trade workers interact with a specific object (e.g., working tools or construction materials) relevant to their trades. This research developed an approach to understand the context from sequential image frames based on four features: posture, object, spatial features, and temporal feature. Both posture and object features were used to analyze the interaction between the worker and the target object, and the other two features were used to detect movements from the entire region of image frames in both temporal and spatial domains. The developed approach used convolutional neural networks (CNN) for feature extractors and activity classifiers and long short-term memory (LSTM) was also used as an activity classifier. The developed approach provided an average accuracy of 85.96% for classifying 12 target construction tasks performed by two trades of workers, which was higher than two benchmark models. This experimental result indicated that integrating a concept of the human-object interaction offers great benefits in activity recognition when various trade workers coexist in a scene.

• Advances in concrete construction
• /
• v.14 no.3
• /
• pp.185-194
• /
• 2022

For producing cement and concrete, the construction field has been encouraged by the usage of industrial soil waste (or) secondary materials since it decreases the utilization of natural resources. Simultaneously, for ensuring the quality, the analyses of the strength along with durability properties of that sort of cement and concrete are required. The prediction of strength along with other properties of High-Performance Concrete (HPC) by optimization and machine learning algorithms are focused by already available research methods. However, an error and accuracy issue are possessed. Therefore, the Enhanced Deep Neural Network (EDNN) based strength along with durability prediction of HPC was utilized by this research method. Initially, the data is gathered in the proposed work. Then, the data's pre-processing is done by the elimination of missing data along with normalization. Next, from the pre-processed data, the features are extracted. Hence, the data input to the EDNN algorithm which predicts the strength along with durability properties of the specific mixing input designs. Using the Switched Multi-Objective Jellyfish Optimization (SMOJO) algorithm, the weight value is initialized in the EDNN. The Gaussian radial function is utilized as the activation function. The proposed EDNN's performance is examined with the already available algorithms in the experimental analysis. Based on the RMSE, MAE, MAPE, and R² metrics, the performance of the proposed EDNN is compared to the existing DNN, CNN, ANN, and SVM methods. Further, according to the metrices, the proposed EDNN performs better. Moreover, the effectiveness of proposed EDNN is examined based on the accuracy, precision, recall, and F-Measure metrics. With the already-existing algorithms i.e., JO, GWO, PSO, and GA, the fitness for the proposed SMOJO algorithm is also examined. The proposed SMOJO algorithm achieves a higher fitness value than the already available algorithm.