• Proceedings of the Korean Society of Crop Science Conference
• /
• 2022.10a
• /
• pp.80-80
• /
• 2022

Estimation of heading date of paddy rice is laborious and time consuming. Therefore, automatic estimation of heading date of paddy rice is highly essential. In this experiment, deep learning classification models were used to classify two difference categories of rice (vegetative and reproductive stage) based on the panicle initiation of paddy field. Specifically, the dataset includes 444 slanted view images belonging to two categories and was then expanded to include 1,497 images via IMGAUG data augmentation technique. We adopt two transfer learning strategies: (First, used transferring model weights already trained on ImageNet to six classification network models: VGGNet, ResNet, DenseNet, InceptionV3, Xception and MobileNet, Second, fine-tuned some layers of the network according to our dataset). After training the CNN model, we used several evaluation metrics commonly used for classification tasks, including Accuracy, Precision, Recall, and F1-score. In addition, GradCAM was used to generate visual explanations for each image patch. Experimental results showed that the InceptionV3 is the best performing model in terms of the accuracy, average recall, precision, and F1-score. The fine-tuned InceptionV3 model achieved an overall classification accuracy of 0.95 with a high F1-score of 0.95. Our CNN model also represented the change of rice heading date under different date of transplanting. This study demonstrated that image based deep learning model can reliably be used as an automatic monitoring system to detect the heading date of rice crops using CCTV camera.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.6
• /
• pp.39-51
• /
• 2024

This paper proposes a novel approach to enhance the security of container-based systems by analyzing system calls to dynamically detect race conditions without modifying the kernel. Container escape attacks allow attackers to break out of a container's isolation and access other systems, utilizing vulnerabilities such as race conditions that can occur in parallel computing environments. To effectively detect and defend against such attacks, this study utilizes eBPF to observe system call patterns during attack attempts and employs a AdaBoost model to detect them. For this purpose, system calls invoked during the attacks such as Dirty COW and Dirty Cred from popular applications such as MongoDB, PostgreSQL, and Redis, were used as training data. The experimental results show that this method achieved a precision of 99.55%, a recall of 99.68%, and an F1-score of 99.62%, with the system overhead of 8%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 1999.05a
• /
• pp.50-54
• /
• 1999

This paper is proposed to image retrieval system using color and shape feature. Color feature used to four maximum value feature among the maximum value extracted from local color distribution histogram. The preprocessing of shape feature consist of edge extraction and weight central point extraction and angular sampling. The sum of distance from weight central point to contour and variation and max/min used to shape feature. The similarity is estimated compare feature of query image with the feature of images in database and the candidate of image is retrieved in order of similarity. We evaluate the effectiveness of shape feature and color feature in experiment used to two hundred of the closed image. The Recall and the Precision is each 0.72 and 0.53 in the result of average experiment. So the proposed method is presented useful method.

• Korean Journal of Remote Sensing
• /
• v.38 no.6_2
• /
• pp.1677-1689
• /
• 2022

Air pollution is social issue that has long-term and short-term harmful effect on the health of animals, plants, and environments. Chimneys are the primary source of air pollutants that pollute the atmosphere, so their location and type must be detected and monitored. Power plants and industrial complexes where chimneys emit air pollutants, are much less accessible and have a large site, making direct monitoring cost-inefficient and time-inefficient. As a result, research on detecting chimneys using remote sensing data has recently been conducted. In this study, YOLOv5-based chimney detection model was generated using BUAA-FFPP60 open dataset create for power plants in Hebei Province, Tianjin, and Beijing, China. To improve the detection model's performance, data split and data augmentation techniques were used, and a training strategy was developed for optimal model generation. The model's performance was confirmed using various indicators such as precision and recall, and the model's performance was finally evaluated by comparing it to existing studies using the same dataset.

• Journal of KIISE:Software and Applications
• /
• v.29 no.3
• /
• pp.145-155
• /
• 2002

In content-based image retrieval (CBIR), most conventional approaches assume a linear relationship between different features and require users themselves to assign the appropriate weights to each feature. However, the linear relationship assumed between the features is too restricted to accurately represent high-level concepts and the intricacies of human perception. In this paper, a neural network-based image retrieval (NNIR) model is proposed. It has been developed based on a human-computer interaction approach to CBIR using a radial basis function network (RBFN). By using the RBFN, this approach determines the nonlinear relationship between features and it allows the user to select an initial query image and search incrementally the target images via relevance feedback so that more accurate similarity comparison between images can be supported. The experiment was performed to calculate the level of recall and precision based on a database that contains 1,015 images and consists of 145 classes. The experimental results showed that the recall and level of the proposed approach were 93.45% and 80.61% respectively, which is superior than precision the existing approaches such as the linearly combining approach, the rank-based method, and the backpropagation algorithm-based method.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.44 no.5
• /
• pp.1-11
• /
• 2007

As the availability of an image information has been significantly increasing, necessity of system that can manage an image information is increasing. Accordingly, we proposed the region-based content retrieval(CBIR) algorithm based on an efficient combination of an image segmentation, an image texture, a color feature and an image's shape and position information. As a color feature, a HSI color histogram is chosen which is known to measure spatial of colors well. We used active contour and CWT(complex wavelet transform) to perform an image segmentation and extracting an image texture. And shape and position information are obtained using Hu invariant moments in the luminance of HSI model. For efficient similarity computation, the extracted features(color histogram, Hu invariant moments, and complex wavelet transform) are combined and then precision and recall are measured. As a experimental result using DB that was supported by www.freefoto.com. the proposed image retrieval engine have 94.8% precision, 82.7% recall and can apply successfully image retrieval system.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.45 no.6
• /
• pp.179-186
• /
• 2008

In this paper we analyse the effects of Bayesian algorithm in teaming class imbalance problems and compare the performance evaluation methods. The teaming performance of the Bayesian algorithm is evaluated over the class imbalance problems generated by priori data distribution, imbalance data rate and discrimination complexity. The experimental results are calculated by the AUC(Area Under the Curve) values of both ROC(Receiver Operator Characteristic) and PR(Precision-Recall) evaluation measures and compared according to imbalance data rate and discrimination complexity. In comparison and analysis, the Bayesian algorithm suffers from the imbalance rate, as the same result in the reported researches, and the data overlapping caused by discrimination complexity is the another factor that hampers the learning performance. As the discrimination complexity and class imbalance rate of the problems increase, the learning performance of the AUC of a PR measure is much more variant than that of the AUC of a ROC measure. But the performances of both measures are similar with the low discrimination complexity and class imbalance rate of the problems. The experimental results show 4hat the AUC of a PR measure is more proper in evaluating the learning of class imbalance problem and furthermore gets the benefit in designing the optimal learning model considering a misclassification cost.

• The KIPS Transactions:PartD
• /
• v.13D no.7 s.110
• /
• pp.1027-1038
• /
• 2006

There are many technical problems in the recommendation system based on very large database(VLDB). So, it is necessary to study the recommendation system' structure and the data-mining technique suitable for the large scale Internet shopping mail. Thus we design and implement the product recommendation system using k-means clustering algorithm and sequential pattern technique which can be used in large scale Internet shopping mall. This paper processes user information by batch processing, defines the various categories by hierarchical structure, and uses a sequential pattern mining technique for the search engine. For predictive modeling and experiment, we use the real data(user's interest and preference of given category) extracted from log file of the major Internet shopping mall in Korea during 30 days. And we define PRP(Predictive Recommend Precision), PRR(Predictive Recommend Recall), and PF1(Predictive Factor One-measure) for evaluation. In the result of experiments, the best recommendation time and the best learning time of our system are much as O(N) and the values of measures are very excellent.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.10C
• /
• pp.1414-1424
• /
• 2004

In this paper, an algorithm using wavelet transform for detecting a cut that is a radical scene transition point, and fade and dissolve that are gradual scene transition points is proposed. The conventional methods using wavelet transform for this purpose is using features in both spatial and frequency domain. But in the proposed algorithm, the color space of an input image is converted to YUV and then luminance component Y is transformed in frequency domain using 2-level lifting. Then, the histogram of only low frequency subband that may contain some spatial domain features is compared with the previous one. Edges obtained from other higher bands can be divided into global, semi-global and local regions and the histogram of each edge region is compared. The experimental results show the performance improvement of about 17% in recall and 18% in precision and also show a good performance in fade and dissolve detection.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.4
• /
• pp.78-96
• /
• 2023

Urban traffic information is collected by various systems such as VDS, DSRC, and radar. Recently, with the development of deep learning technology, smart intersection systems are expanding, are more widely distributed, and it is possible to collect a variety of information such as traffic volume, and vehicle type and speed. However, as a result of reviewing related literature, the performance evaluation criteria so far are rbs-based evaluation systems that do not consider the deep learning area, and only consider the percent error of 'reference value-measured value'. Therefore, a new performance evaluation method is needed. Therefore, in this study, individual error, interval error, and overall error are calculated by using a formula that considers deep learning performance indicators such as precision and recall based on data ratio and weight. As a result, error rates for measurement value 1 were 3.99 and 3.54, and rates for measurement value 2 were 5.34 and 5.07.