Article
Agronomy
Xinyu Dong, Qi Wang, Qianding Huang, Qinglong Ge, Kejun Zhao, Xingcai Wu, Xue Wu, Liang Lei, Gefei Hao
Summary: Plant diseases pose a threat to global food security, making the diagnosis of plant diseases crucial for agricultural production. Artificial intelligence, particularly deep learning, has emerged as a promising alternative to traditional diagnosis methods due to their limitations. By utilizing pre-trained models specific to plant diseases, our research enhances the accuracy and efficiency of disease diagnosis tasks, supporting better identification, detection, and segmentation of plant diseases.
Article
Computer Science, Artificial Intelligence
Song Lin, Zhiyong He, Lining Sun
Summary: This paper proposes a Micro-defect classification system based on attention enhancement (MDCS) for solving the detection and classification of micro-defects. The system combines defect detection with defect classification and utilizes attention formation and trilinear feature confluence to reduce noise interference. Experimental results show that the system can accurately detect and localize small objects and significantly improve the classification accuracy of micro-defects.
JOURNAL OF INTELLIGENT MANUFACTURING
(2023)
Article
Engineering, Civil
Pravee Kruachottikul, Nagul Cooharojananone, Gridsada Phanomchoeng, Thira Chavarnakul, Kittikul Kovitanggoon, Donnaphat Trakulwaranont
Summary: This study developed a deep learning-based visual defect inspection system for reinforced concrete bridge substructures, which consists of four main components: image acquisition, defect detection, defect classification, and severity prediction. The system achieved high accuracy rates for defect detection, classification, and severity prediction, which were well-received and deemed suitable for practical application by Thailand's Department of Highways.
JOURNAL OF CIVIL STRUCTURAL HEALTH MONITORING
(2021)
Article
Plant Sciences
Shivali Amit Wagle, R. Harikrishnan, Sawal Hamid Md Ali, Mohammad Faseehuddin
Summary: Precision crop safety can be improved using automated systems for detecting and classifying plants. This study proposes compact convolutional neural networks and AlexNet with transfer learning for plant classification. The proposed models achieve high accuracy and require less training time compared to AlexNet.
Article
Plant Sciences
Chin Poo Lee, Kian Ming Lim, Yu Xuan Song, Ali Alqahtani
Summary: Plant leaf classification involves identifying and categorizing plant species based on leaf characteristics. Machine learning techniques have been used to improve accuracy, but rely on large amounts of training data. To overcome this, a Plant-CNN-ViT ensemble model is proposed, combining the strengths of four pre-trained models.
Article
Engineering, Multidisciplinary
Juyoung Park, Jung Hee Lee, Junseong Bang
Summary: PotholeEye+ is a mobile system that automatically monitors the surface of a roadway and detects pavement distress in real-time through video analysis. Tested on real highways for a year, it showed high accuracy, precision, and recall rates while driving at an average speed of 110 km/h.
CMES-COMPUTER MODELING IN ENGINEERING & SCIENCES
(2021)
Article
Chemistry, Analytical
Yanlong Cao, Binjie Ding, Jingxi Chen, Wenyuan Liu, Pengning Guo, Liuyi Huang, Jiangxin Yang
Summary: Automated inspection technology based on computer vision is widely used in manufacturing to achieve high speed and accuracy. However, the presence of metal parts with high gloss or shadow can cause overexposure in captured images. To address this issue, a photometric-stereo-based defect detection system (PSBDDS) is proposed, which combines photometric stereo with defect detection to eliminate interference from highlights and shadows. Additionally, a photometric-stereo-based defect detection framework is introduced, along with the creation of a photometric stereo defect detection (PSDD) dataset to bridge the gap between learning-based photometric stereo and defect detection methods. Experimental results demonstrate the effectiveness of the proposed PSBBD and PSDD dataset.
Article
Computer Science, Information Systems
Yerin Lee, Soyoung Lim, Il-Youp Kwak
Summary: The study proposed two models to tackle the challenges faced by acoustic scene classification, including handling audio recorded using different devices and developing a low-complexity model.
Article
Computer Science, Information Systems
Okeke Stephen, Uchenna Joseph Maduh, Mangal Sain
Summary: We propose a simple yet effective convolutional neural network for learning the similarities between closely related raw pixel images and extracting feature representations for classification. The network initializes convolutional kernels from learned filter kernels, and simultaneously learns binary-class classification of sigmoid and discriminative feature vectors. Compared to traditional handcrafted feature extraction methods, which separate feature extraction and classification tasks during training, our method achieves efficient classification using high-quality feature representations learned by the network. Experimental evaluation on tile surface images with cracked and no-cracked surfaces demonstrates the effectiveness of our proposed method for automated visual inspections of surface defects.
Article
Computer Science, Artificial Intelligence
Omar Bin Samin, Maryam Omar, Musadaq Mansoor
Summary: The study introduces a deep learning architecture model, CapPlant, for predicting plant health or disease from images with significant improvements in prediction accuracy. Tested on the PlantVillage dataset, the model achieved an overall test accuracy of 93.01% and an F1 score of 93.07%.
PEERJ COMPUTER SCIENCE
(2021)
Article
Computer Science, Interdisciplinary Applications
Kung-Jeng Wang, Hao Fan-Jiang, Ya-Xuan Lee
Summary: This study proposes a four-stage defect detection model using convolution neural networks (CNNs) to reduce error rates and offer informative quality messages. The model achieves high precision and velocity and reduces labor power.
COMPUTERS & INDUSTRIAL ENGINEERING
(2022)
Article
Computer Science, Information Systems
Hemanta Kumar Bhuyan, A. Vijayaraj, Vinayakumar Ravi
Summary: This paper discusses a single setting framework for the diagnosis system of cancer disease. It focuses on utilizing a convolutional neural network (CNN) architecture with deep learning approaches to determine the relevant illness of patients through affected images. The proposed model outperforms existing models in terms of accuracy and area under the curve (AUC), demonstrating its effectiveness in conventional detection, segmentation, and classification methods. The suggested diagnostics method can provide support for radiologists in each stage of image processing of the infected region.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Agronomy
Kishore K. Kumar, E. Kannan
Summary: Research on using machine learning and plant images for detecting plant diseases has significantly increased in agriculture. This paper presents a prototype for detecting diseases in rice plants, such as bacterial leaf blight, brown spot, and leaf smut. The proposed prototype utilizes image processing and machine learning algorithms to extract important features from diseased rice leaf images, achieving an accuracy rate of approximately 98.8% in disease detection and classification.
Review
Agronomy
Jinzhu Lu, Lijuan Tan, Huanyu Jiang
Summary: This review article discusses the latest CNN networks relevant to plant leaf disease classification, summarizes the DL principles involved in plant disease classification, presents the main problems and corresponding solutions of CNN used for plant disease classification, and discusses the future development direction in plant disease classification.
Article
Engineering, Multidisciplinary
Qiang Li, Qinyuan Huang, Tian Yang, Ying Zhou, Kun Yang, Hong Song
Summary: This paper presents an improved convolutional neural network (CNN) model based on multi-head attention for detecting internal defects in arc magnets. The model efficiently highlights important features and accelerates the training process using the characteristics of multi-head attention. A data augmentation method is designed to meet the data requirements of the model. Experimental results demonstrate that the approach has superior inspection performance in different test scenarios.