• Journal of Information Processing Systems
• /
• v.20 no.2
• /
• pp.149-158
• /
• 2024

Traditional manual identification of crop leaf diseases is challenging. Owing to the limitations in manpower and resources, it is challenging to explore crop diseases on a large scale. The emergence of artificial intelligence technologies, particularly the extensive application of deep learning technologies, is expected to overcome these challenges and greatly improve the accuracy and efficiency of crop disease identification. Crop leaf disease identification models have been designed and trained using large-scale training data, enabling them to predict different categories of diseases from unlabeled crop leaves. However, these models, which possess strong feature representation capabilities, require substantial training data, and there is often a shortage of such datasets in practical farming scenarios. To address this issue and improve the feature learning abilities of models, this study proposes a deep transfer learning adaptation strategy. The novel proposed method aims to transfer the weights and parameters from pre-trained models in similar large-scale training datasets, such as ImageNet. ImageNet pre-trained weights are adopted and fine-tuned with the features of crop leaf diseases to improve prediction ability. In this study, we collected 16,060 crop leaf disease images, spanning 12 categories, for training. The experimental results demonstrate that an impressive accuracy of 98% is achieved using the proposed method on the transferred ResNet-50 model, thereby confirming the effectiveness of our transfer learning approach.

• Smart Structures and Systems
• /
• v.33 no.4
• /
• pp.253-262
• /
• 2024

Current-based mechanical fault diagnosis is more convenient and low cost since additional sensors are not required. However, it is still challenging to achieve this goal due to the weak fault information in current signals. In this paper, a dual-loss convolutional neural network (DLCNN) is proposed to implement the intelligent bearing fault diagnosis via current signals. First, a novel similarity loss (SimL) function is developed, which is expected to maximize the intra-class similarity and minimize the inter-class similarity in the model optimization operation. In the loss function, a weight parameter is further introduced to achieve a balance and leverage the performance of SimL function. Second, the DLCNN model is constructed using the presented SimL and the cross-entropy loss. Finally, the two-phase current signals are fused and then fed into the DLCNN to provide more fault information. The proposed DLCNN is tested by experiment data, and the results confirm that the DLCNN achieves higher accuracy compared to the conventional CNN. Meanwhile, the feature visualization presents that the samples of different classes are separated well.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.1
• /
• pp.35-43
• /
• 2010

1D PCA utilized in the face appearance-based face recognition methods such as eigenface-based face recognition method may lead to less face representative power and more computational cost due to the resulting 1D face appearance data vector of high dimensionality. To resolve such problems of 1D PCA, 2D PCA-based face recognition methods had been developed. However, the face representation model obtained by direct application of 2D PCA to a face image set includes both face common features and face distinctive identity features. Face common features not only prevent face recognizability but also cause more computational cost. In this paper, we first develope a model of a face distinctive identity feature subspace separated from the effects of face common features in the face feature space obtained by application of 2D PCA analysis. Then, a novel robust face recognition based on the face distinctive identity feature subspace model is proposed. The proposed face recognition method based on the face distinctive identity feature subspace shows better performance than the conventional PCA-based methods (1D PCA-based one and 2D PCA-based one) with respect to recognition rate and processing time since it depends only on the face distinctive identity features. This is verified through various experiments using Yale A and IMM face database consisting of face images with various face poses under various illumination conditions.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.205-225
• /
• 2018

Convolutional Neural Network (ConvNet) is one class of the powerful Deep Neural Network that can analyze and learn hierarchies of visual features. Originally, first neural network (Neocognitron) was introduced in the 80s. At that time, the neural network was not broadly used in both industry and academic field by cause of large-scale dataset shortage and low computational power. However, after a few decades later in 2012, Krizhevsky made a breakthrough on ILSVRC-12 visual recognition competition using Convolutional Neural Network. That breakthrough revived people interest in the neural network. The success of Convolutional Neural Network is achieved with two main factors. First of them is the emergence of advanced hardware (GPUs) for sufficient parallel computation. Second is the availability of large-scale datasets such as ImageNet (ILSVRC) dataset for training. Unfortunately, many new domains are bottlenecked by these factors. For most domains, it is difficult and requires lots of effort to gather large-scale dataset to train a ConvNet. Moreover, even if we have a large-scale dataset, training ConvNet from scratch is required expensive resource and time-consuming. These two obstacles can be solved by using transfer learning. Transfer learning is a method for transferring the knowledge from a source domain to new domain. There are two major Transfer learning cases. First one is ConvNet as fixed feature extractor, and the second one is Fine-tune the ConvNet on a new dataset. In the first case, using pre-trained ConvNet (such as on ImageNet) to compute feed-forward activations of the image into the ConvNet and extract activation features from specific layers. In the second case, replacing and retraining the ConvNet classifier on the new dataset, then fine-tune the weights of the pre-trained network with the backpropagation. In this paper, we focus on using multiple ConvNet layers as a fixed feature extractor only. However, applying features with high dimensional complexity that is directly extracted from multiple ConvNet layers is still a challenging problem. We observe that features extracted from multiple ConvNet layers address the different characteristics of the image which means better representation could be obtained by finding the optimal combination of multiple ConvNet layers. Based on that observation, we propose to employ multiple ConvNet layer representations for transfer learning instead of a single ConvNet layer representation. Overall, our primary pipeline has three steps. Firstly, images from target task are given as input to ConvNet, then that image will be feed-forwarded into pre-trained AlexNet, and the activation features from three fully connected convolutional layers are extracted. Secondly, activation features of three ConvNet layers are concatenated to obtain multiple ConvNet layers representation because it will gain more information about an image. When three fully connected layer features concatenated, the occurring image representation would have 9192 (4096+4096+1000) dimension features. However, features extracted from multiple ConvNet layers are redundant and noisy since they are extracted from the same ConvNet. Thus, a third step, we will use Principal Component Analysis (PCA) to select salient features before the training phase. When salient features are obtained, the classifier can classify image more accurately, and the performance of transfer learning can be improved. To evaluate proposed method, experiments are conducted in three standard datasets (Caltech-256, VOC07, and SUN397) to compare multiple ConvNet layer representations against single ConvNet layer representation by using PCA for feature selection and dimension reduction. Our experiments demonstrated the importance of feature selection for multiple ConvNet layer representation. Moreover, our proposed approach achieved 75.6% accuracy compared to 73.9% accuracy achieved by FC7 layer on the Caltech-256 dataset, 73.1% accuracy compared to 69.2% accuracy achieved by FC8 layer on the VOC07 dataset, 52.2% accuracy compared to 48.7% accuracy achieved by FC7 layer on the SUN397 dataset. We also showed that our proposed approach achieved superior performance, 2.8%, 2.1% and 3.1% accuracy improvement on Caltech-256, VOC07, and SUN397 dataset respectively compare to existing work.

• Journal of Digital Convergence
• /
• v.13 no.6
• /
• pp.151-156
• /
• 2015

In this paper, we have proposed personal multimodal biometric authentication system based on face detection, recognition and speaker verification for smart-phone environment. Proposed system detect the face with Modified Census Transform algorithm then find the eye position in the face by using gabor filter and k-means algorithm. Perform preprocessing on the detected face and eye position, then we recognize with Linear Discriminant Analysis algorithm. Afterward in speaker verification process, we extract the feature from the end point of the speech data and Mel Frequency Cepstral Coefficient. We verified the speaker through Dynamic Time Warping algorithm because the speech feature changes in real-time. The proposed multimodal biometric system is to fuse the face and speech feature (to optimize the internal operation by integer representation) for smart-phone based real-time face detection, recognition and speaker verification. As mentioned the multimodal biometric system could form the reliable system by estimating the reasonable performance.

• Journal of the Korean Society for information Management
• /
• v.22 no.3 s.57
• /
• pp.261-287
• /
• 2005

The purpose of this paper is to explore the ways to improve the performance of SVM (Support Vector Machines) text classifier using inter-document similarities. SVMs are powerful machine learning systems, which are considered as the state-of-the-art technique for automatic document classification. In this paper text categorization via SVMs approach based on feature representation with document vectors is suggested. In this approach, document vectors instead of index terms are used as features, and vector similarities instead of term weights are used as feature values. Experiments show that SVM classifier with document vector features can improve the document classification performance. For the sake of run-time efficiency, two methods are developed: One is to select document vector features, and the other is to use category centroid vector features instead. Experiments on these two methods show that we can get improved performance with small vector feature set than the performance of conventional methods with index term features.

• Spatial Information Research
• /
• v.19 no.5
• /
• pp.75-86
• /
• 2011

Three-dimensional modeling of cities in the real-world is an essential task for city planning and decision-making. And many three-dimensional city models are being developed with the development of wireless Internet and location-based services that identify the location of users and provide the information increases for consumers. Especially, in case of urban areas of Korea, indoor space modeling as well as outdoor is needed due to the high-rise buildings densities. Also location-based services should be provided through spatial analysis such as the shortest path based on a space model. Many studies of three-dimensional city models are feature models. In a feature model, space is represented by combining primitives, and relationships among spaces are represented only if shared primitives are detected. So relationships between complex three-dimensional objects in space is difficult to be defined through the feature models. In this study, Navigable space-relation model(NSRM) is developed, which is topological data model for efficient representation of spatial relationships between objects based on the network structure.

• Nuclear Engineering and Technology
• /
• v.52 no.7
• /
• pp.1429-1435
• /
• 2020

Applying an accurate parametric prediction model to identify abnormal or false pressurizer water levels (PWLs) is critical to the safe operation of marine pressurized water reactors (PWRs). Recently, deep-learning-based models have proved to be a powerful feature extractor to perform high-accuracy prediction. However, the effectiveness of models still suffers from two issues in PWL prediction: the correlations shifting over time between PWL and other feature parameters, and the example imbalance between fluctuation examples (minority) and stable examples (majority). To address these problems, we propose a cost-sensitive mechanism to facilitate the model to learn the feature representation of later examples and fluctuation examples. By weighting the standard mean square error loss with a cost-sensitive factor, we develop a Cost-Sensitive Long Short-Term Memory (CSLSTM) model to predict the PWL of PWRs. The overall performance of the CSLSTM is assessed by a variety of evaluation metrics with the experimental data collected from a marine PWR simulator. The comparisons with the Long Short-Term Memory (LSTM) model and the Support Vector Regression (SVR) model demonstrate the effectiveness of the CSLSTM.

• Journal of Fashion Business
• /
• v.24 no.3
• /
• pp.101-113
• /
• 2020

This study considers aesthetic characteristics by examining the fashion style of the silver fashion icon Iris Apfel. The research methods were a quantitative and qualitative analysis of Iris Apfel's images that were collected from 2015 to 2019 on various web-sites according to four criterions following advanced research analysis of fashion style. The results of the study are as follows. The analysis results on the fashion style of Iris Apfel, an icon of silver style, showed that cocoon, barrel and A-line silhouettes appeared most in terms of silhouettes. Second, in terms of colors, achromatic colors dominated among solid colors while one particular vivid color appeared most it came to mixed color. In terms of multi colors, these appeared according to the patterns applied to her clothing, in particular, colorful colors were used to emphasize splendor. Third, flower, bird and geometric patterns appeared most in terms of material patterns. Lastly, it was found that white short cut hair, large necklaces or bangle bracelets, over-sized black glasses and fur mufflers or canes were used in terms of hair and accessories. The features derived through analysis of the fashion style of Iris Apfel, an icon of silver style, are as follows. The first feature is exaggeration through splendid primary colors and over-sized silhouettes. The second feature is the hybrid of modern composition methods using natural images and exotic preferences. The third feature is her representation of identity using fixed items.

• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.11
• /
• pp.541-548
• /
• 2016

The use of social media is already essential as a source of information for listening user's various opinions and monitoring. We define social 'risks' that issues effect negative influences for public opinion in social media. This paper aims to discriminate various linguistic features and reveal their effects for building an automatic classification model of social risks. Expecially we adopt a word embedding technique for representation of linguistic clues in risk sentences. As a preliminary experiment to analyze characteristics of individual features, we revise errors in automatic linguistic analysis. At the result, the most important feature is NE (Named Entity) information and the best condition is when combine basic linguistic features. word embedding, and word clusters within core predicates. Experimental results under the real situation in social bigdata - including linguistic analysis errors - show 92.08% and 85.84% in precision respectively for frequent risk categories set and full test set.