이 연구에서는 Inception V3, SqueezeNet(local), VGG-16, Painters 및 DeepLoc의 다섯 가지 인공지능(AI) 모 델을 사용하여 차나무 잎의 병해를 분류하였다. 여덟 가지 이미지 카테고리를 사용하였는데, healthy, algal leaf spot, anthracnose, bird’s eye spot, brown blight, gray blight, red leaf spot, and white spot였다. 이 연구에서 사용한 소프트웨 어는 데이터 시각적 프로그래밍을 위한 파이썬 라이브러리로 작동하는 Orange3였다. 이는 데이터를 시각적으로 조작하여 분석하기 위한 워크플로를 생성하는 인터페이스를 통해 작동되었다. 각 AI 모델의 정확도로 최적의 AI 모 델을 선택하였다. 모든 모델은 Adam 최적화, ReLU 활성화 함수, 은닉 레이어에 100개의 뉴런, 신경망의 최대 반복 횟수가 200회, 그리고 0.0001 정규화를 사용하여 훈련되었다. Orange3 기능을 확장하기 위해 새로운 이미지 분석 Add-on을 설치하였다. 훈련 모델에서는 이미지 가져오기(import image), 이미지 임베딩(image embedding), 신경망 (neural network), 테스트 및 점수(test and score), 혼동 행렬(confusion matrix) 위젯이 사용되었으며, 예측에는 이미 지 가져오기(import image), 이미지 임베딩(image embedding), 예측(prediction) 및 이미지 뷰어(image viewer) 위젯 이 사용되었다. 다섯 AI 모델[Inception V3, SqueezeNet(로컬), VGG-16, Painters 및 DeepLoc]의 신경망 정밀도는 각 각 0.807, 0.901, 0.780, 0.800 및 0.771이었다. 결론적으로 SqueezeNet(local) 모델이 차나무 잎 이미지를 사용하여 차 병해 탐색을 위한 최적 AI 모델로 선택되었으며, 정확도와 혼동 행렬을 통해 뛰어난 성능을 보였다.

Recently, transfer learning techniques with a base convolutional neural network (CNN) model have widely gained acceptance in early detection and classification of crop diseases to increase agricultural productivity with reducing disease spread. The transfer learning techniques based classifiers generally achieve over 90% of classification accuracy for crop diseases using dataset of crop leaf images (e.g., PlantVillage dataset), but they have ability to classify only the pre-trained diseases. This paper provides with an evaluation scheme on selecting an effective base CNN model for crop disease transfer learning with regard to the accuracy of trained target crops as well as of untrained target crops. First, we present transfer learning models called CDC (crop disease classification) architecture including widely used base (pre-trained) CNN models. We evaluate each performance of seven base CNN models for four untrained crops. The results of performance evaluation show that the DenseNet201 is one of the best base CNN models.

This research examines deep learning based image recognition models for beef sirloin classification. The sirloin of beef can be classified as the upper sirloin, the lower sirloin, and the ribeye, whereas during the distribution process they are often simply unified into the sirloin region. In this work, for detailed classification of beef sirloin regions we develop a model that can learn image information in a reasonable computation time using the MobileNet algorithm. In addition, to increase the accuracy of the model we introduce data augmentation methods as well, which amplifies the image data collected during the distribution process. This data augmentation enables to consider a larger size of training data set by which the accuracy of the model can be significantly improved. The data generated during the data proliferation process was tested using the MobileNet algorithm, where the test data set was obtained from the distribution processes in the real-world practice. Through the computational experiences we confirm that the accuracy of the suggested model is up to 83%. We expect that the classification model of this study can contribute to providing a more accurate and detailed information exchange between suppliers and consumers during the distribution process of beef sirloin.

Background: Only 2% of falls in older adults result in serious injuries (i.e., hip fracture). Therefore, it is important to differentiate injurious versus non-injurious falls, which is critical to develop effective interventions for injury prevention. Objects: The purpose of this study was to a. extract the best features of surface electromyography (sEMG) for classification of injurious falls, and b. find a best model provided by data mining techniques using the extracted features. Methods: Twenty young adults self-initiated falls and landed sideways. Falling trials were consisted of three initial fall directions (forward, sideways, or backward) and three knee positions at the time of hip impact (the impacting-side knee contacted the other knee (“knee together”) or the mat (“knee on mat”), or neither the other knee nor the mat was contacted by the impacting-side knee (“free knee”). Falls involved “backward initial fall direction” or “free knee” were defined as “injurious falls” as suggested from previous studies. Nine features were extracted from sEMG signals of four hip muscles during a fall, including integral of absolute value (IAV), Wilson amplitude (WAMP), zero crossing (ZC), number of turns (NT), mean of amplitude (MA), root mean square (RMS), average amplitude change (AAC), difference absolute standard deviation value (DASDV). The decision tree and support vector machine (SVM) were used to classify the injurious falls. Results: For the initial fall direction, accuracy of the best model (SVM with a DASDV) was 48%. For the knee position, accuracy of the best model (SVM with an AAC) was 49%. Furthermore, there was no model that has sensitivity and specificity of 80% or greater. Conclusion: Our results suggest that the classification model built upon the sEMG features of the four hip muscles are not effective to classify injurious falls. Future studies should consider other data mining techniques with different muscles.

PURPOSES : This study uses deep learning image classification models and vehicle-mounted cameras to detect types of pavement distress — such as potholes, spalling, punch-outs, and patching damage — which require urgent maintenance. METHODS : For the automatic detection of pavement distress, the optimal mount location on a vehicle for a regular action camera was first determined. Using the orthogonal projection of obliquely captured surface images, morphological operations, and multi-blob image processing, candidate distressed pavement images were extracted from road surface images of a 16,036 km in-lane distance. Next, the distressed pavement images classified by experts were trained and tested for evaluation by three deep learning convolutional neural network (CNN) models: GoogLeNet, AlexNet, and VGGNet. The CNN models were image classification tools used to identify and extract the combined features of the target images via deep layers. Here, a data augmentation technique was applied to produce big distress data for training. Third, the dimensions of the detected distressed pavement patches were computed to estimate the quantity of repair materials needed. RESULTS : It was found that installing cameras 1.8 m above the ground on the exterior rear of the vehicle could provide clear pavement surface images with a resolution of 1 cm per pixel. The sensitivity analysis results of the trained GoogLeNet, AlexNet, and VGGNet models were 93 %, 86 %, and 72 %, respectively, compared to 62.7 % for the dimensional computation. Following readjustment of the image categories in the GoogLeNet model, distress detection sensitivity increased to 94.6 %. CONCLUSIONS : These findings support urgent maintenance by sending the detected distressed pavement images with the dimensions of the distressed patches and GPS coordinates to local maintenance offices in real-time.

국내에는 독자적으로 연구가 수행되어 개인적으로 보관 중인 지질 연구 자료가 다량 존재하는데, 이 자료에 대한 접근성이 떨어지기 때문에 다른 연구자들과의 공유가 용이하지 않다. 이런 자료에 대한 메타데이터를 체계적으로 구축하고 총괄적으로 관리하여 이 자료를 필요로 하는 연구자들이 효과적으로 연구를 수행할 수 있는 기회를 제공하는 것이 이 연구의 목적이다. 국내에서 연구된 약 1000여개의 지질 시료(900여개의 암석과 화석 시료, 100여개의 박편 시 료)를 수집하였고, 각 시료의 고화질 사진, 분류, 시료명, 보유기관, 산지, 좌표, 특징 등에 대한 메타데이터를 구축하였다. 암석과 화석 시료 100개에 대해 추가적으로 3D 모델링을 수행하였다. 이 연구를 통해 유실되거나 방치되는 중요한 지질 자료에 대한 연구자들의 접근성이 높아지고 자료의 공유가 가능해진다. 따라서 연구자들은 반복적인 연구 자료 수 집 작업으로 인한 시간과 비용의 낭비를 줄일 수 있고, 효율적인 연구를 수행하여 경쟁력을 갖춘 연구 결과를 획득할 수 있다. 또한 이미 확보된 시료에 대한 무분별한 반복 채집으로 인해 중요한, 그리고 피해에 취약한 자료가 훼손되는 것을 방지할 수 있다. 향후 전국의 대학과 연구기관에서 보관중인 다양한 암석과 박편 시료에 대한 메타데이터를 추가로 구축하면 자료의 식별 및 진전된 연구가 가능하고, 더불어 전문적인 광물학 및 암석학의 기초 지식에 대한 비교와 분석을 기대할 수 있다.

Hidden Markov Models(HMM)을 이용한 생물음향 분석은 장기간 동안 녹음된 대용량 데이터에서 생물종의 식별을 용이하게 수행할 수 있는 머신러닝 기법 중 하나이다. 본 연구에서는 HMM을 이용하여 양서류 번식울음에 대한 자동식별을 목적으로 진행하였다. 연구대상지는 서울시 북한산국립공원 진관동습지이었다. 연구기간은 2018년 6~7월 중 양서류가 집중적으로 번식울음을 내는 3일 선정하였다. 연구대상종은 진관동 습지에서 여름철 번식울음을 내는 청개구리, 무당개구리, 맹꽁이 3종이었다. 번식울음 녹음은 생물음향 측정기기(SM4, Wildlife Acoustics Inc., Concord, MA, USA)를 습지를 바라보도록 수변의 수목에 설치(높이 1.0m)하고, 시간당 5분씩 24시간 녹음되도록 스케쥴을 설정하였다. 양서류 번식울음 자동식별을 위한 분석프로그램은 Kaleidoscope Pro(Wildlife Acoustics Inc., Concord, MA, USA)를 이용하였다. 연구결과, 양서류 종별 번식울음 특성을 고려하여 주파수 범위와 음절 지속시간, 음절 간격을 달리한 결과 분류 정확도가 매우 높은 것으로 나타났다. 청개구리의 경우 주파수 범위를 3000~3600Hz, 지속시간을 0.1초, 음간격을 0.12초 로 설정한 결과 총 45개의 클러스터가 자동분류되었다. 이 중 연구자에 의해 청개구리 번식울음으로 명확히 분류된 클러스터는 45개 중 10개 클러스터였다. 10개 이외의 클러스터는 대부분 야생조류 종이 포함되어 있었다. 무당개구리의 경우 주파수 범위를 500~700Hz, 지속시간을 0.15초, 음 간격을 0.3초로 설정한 결과 무당개구리 이외에 다른 종은 포함되지 않는 단일 클러스터로 분석결과가 도출되었다. 맹꽁이의 경우 주파수 범위를 1500~2500Hz, 지속시간을 0.3초, 음간격을 0.5초로 설정한 결과 맹꽁이 이외에 다른 종은 포함되지 않는 단일 클러스터로 분석결과가 도출되었다. 본 연구는 머신러닝 기법을 이용하여 생물음향 빅데이터를 효율적으로 분석함으로서 생물음향을 이용한 생태계 다양성 평가의 기초자료로 활용이 가능할 것으로 판단된다.

The purpose of this study is to develop the Generalized Depreciation Function (GDF) and Winfrey Depreciation Function (WDF) by reviewing methods for the depreciation accountings. The Depreciation Accounting Models (DAM), including straight-line model, declining-balance model, sum-of-the-year-digit model and sinking fund model presented in this paper, are reclassified into the charging pattern of increasing type, decreasing type and constant type. This paper also discusses the development of the GDFs based on convex type, concave type and constant type according to the demand pattern of product, frequency of plant usage, deterioration of time, relative inadequacy, Capital Expenditure (CAPEX) and Operating Expenditure (OPEX) of the Total Productive Maintenance (TPM). The WDFs presented in this paper depict a sudden degradation of plant performance by measuring the change of TPM activity at the midpoint of useful life of asset. The WDFs are classified into left-modal type, symmetrical type and right-modal type by varying the value of skewness and kurtosis. Moreover, three increasing patterns, such as convex, concave and linear types, are used in this paper to present the distinct identification of WFDs by using Instantaneous Depreciation Rate (IDR) in terms of Performance Depreciation Function (PDF) and Depreciation Density Function (DDF). In order to have better understanding of depreciation models, the numerical examples are used for evaluating the Net Operating Less Adjusted Tax (NOPLAT) and Economic Value Added (EVA). It is concluded that the depreciation models showing a large dispersion of EVA require the adjustment of NOPLAT and Invested Capital (IC) based on the objective cash basis and net operating activity for reducing the variation of EVA.

The study interprets each of three classification models based on Bath-Tub Failure Rate (BTFR), Extreme Value Distribution (EVD) and Conjugate Bayesian Distribution (CBD). The classification model based on BTFR is analyzed by three failure patterns of decreasing, constant, or increasing which utilize systematic management strategies for reliability of time. Distribution model based on BTFR is identified using individual factors for each of three corresponding cases. First, in case of using shape parameter, the distribution based on BTFR is analyzed with a factor of component or part number. In case of using scale parameter, the distribution model based on BTFR is analyzed with a factor of time precision. Meanwhile, in case of using location parameter, the distribution model based on BTFR is analyzed with a factor of guarantee time. The classification model based on EVD is assorted into long-tailed distribution, medium-tailed distribution, and short-tailed distribution by the length of right-tail in distribution, and depended on asymptotic reliability property which signifies skewness and kurtosis of distribution curve. Furthermore, the classification model based on CBD is relied upon conjugate distribution relations between prior function, likelihood function and posterior function for dimension reduction and easy tractability under the occasion of Bayesian posterior updating.