Aug 2026· Foods· Vol 15, pp. 2910· 0 citations· 56 references
Medicine
TL;DR
This study demonstrates the feasibility of CNN-based brown rice defect detection, with future work directed toward lightweight deployment and multimodal fusion for production-line application.
Abstract
In order to enable accurate and efficient rice quality evaluation through full-surface defect detection of brown rice, a detection system based on convolutional neural network (CNN) was developed. A dataset of images of five categories—unhulled, normal, broken, cracked, and insect-bitten brown rice—was collected. Three CNN models, YOLOv5s, YOLOv7, and Faster R-CNN, were evaluated and compared with traditional algorithms including support vector machine (SVM) and back propagation (BP) neural networks. Experimental results showed that CNN-based methods in the present database significantly outperformed traditional approaches, with the YOLOv5s model achieving the best comprehensive performance: 95.80% detection accuracy, 10.90 ms inference time per image, and 92 frames/s processing speed. An improved Rice-YOLOv5s algorithm was further proposed and validated through batch detection experiments, achieving an average recognition accuracy of 96.44% and a processing time of 9.2 ms per image, which is equivalent to approximately 108.7 FPS. This study demonstrates the feasibility of CNN-based brown rice defect detection, with future work directed toward lightweight deployment and multimodal fusion for production-line application.
Effective egg grading and surface defect detection are critical for ensuring operational efficiency, food safety, and quality control in the poultry industry. Conventional manual inspection is labor-intensive, subjective, and impractical for high-throughput environments, while mechanical grading systems often overlook external defects in this domain. This study proposes a machine vision–based framework employing multiple learning strategies, including a custom Convolutional Neural Network (CNN), transfer-learning-based VGG16, YOLOv8n object detection, Random Forest (RF), and a novel hybrid VGG16 and RF model for automated egg grading and crack detection. Experiments were conducted using two datasets: a public egg crack detection dataset and a self-collected RGB dataset that captures real-world variations in lighting, orientation, background, and imaging devices. Eleven image pre-processing pipelines were systematically evaluated to examine their influence on accuracy, robustness, and computational efficiency. Results demonstrate that VGG16 achieved the highest classification accuracy, reaching 98.85% on the public dataset and 97.54% on the self-collected dataset. However, the proposed hybrid VGG16–RF model attained competitive performance (97.91% and 95.25%, respectively) while significantly reducing training and processing time. These findings indicate that the hybrid approach offers a favorable accuracy–efficiency trade-off, making it particularly suitable for real-time or resource-constrained deployment in practical poultry production settings.
Xin-Yao Wong, P. Goh, Ying-Chong Lee et al.· International Conference on...· 0 citations
The development of artificial intelligence increases the risk of facial image manipulation through deepfake technology that is difficult to distinguish visually. This study aims to implement a Convolutional Neural Network (CNN) based on EfficientNetB0 to classify real and deepfake static facial images and integrate the model into a website application. The study follows the CRISP-DM framework using the FaceForencis++ extracted frames dataset from Kaggle, consisting of 60,000 images with 30,000 real images and 30,000 fake images. The data were divided into 70:15:15 ratios for training, validation, and testing, then processed through 224 x 224 pixel resizing, data augmentation, and facial area cropping using OpenCV Haar Cascade during inference. The model achieved 82.50% validation accuracy and 0.3874 validation loss at the 10th epoch. Decision threshold optimization at 0.78 produced 80.99% accuracy, 80.98% macro F1-score, and 78.56% recall for the fake class on the test data. The model was deployed in a Streamlit application, allowing users to upload facial images and receive detection results directly. The results indicate that the combination of EfficientNetB0 and threshold optimization can support preliminary verification of digital facial image authenticity.
Dimas Aditya Nugraha, Syariful Alam, Chandra Dewi Lestari· Jurnal Teknologi Informatika...· 0 citations
Biomass briquettes are increasingly recognized as a promising source of renewable energy, as they are produced from organic waste materials such as agricultural residues, wood scraps, and coconut shells. As an environmentally friendly energy alternative, ensuring their quality is essential to achieve consistent combustion and energy efficiency. Previous studies have explored the evaluation and classification of briquettes using image-based machine learning techniques. One study applied a Convolutional Neural Network (CNN) to distinguish good from bad briquettes, achieving very high performance, with accuracy, precision, and recall reportedly reaching up to 100%. However, these results were obtained using a dataset of briquette images captured on a plain white background with minimal variation. When tested on more complex data images of briquettes against non-uniform backgrounds containing multiple objects, the CNN model’s performance dropped drastically. Under these real-world conditions, accuracy and precision fell below 50%, indicating that the model lacked robustness for practical applications. To overcome this limitation, further research focused on developing an improved classification model using the Region-Based Convolutional Neural Network (R-CNN) method. R-CNN employs a region proposal algorithm using a selective search approach, enhancing object detection and localization in cluttered backgrounds. Experimental testing through a developed interface demonstrated a significant improvement, with accuracy increasing to approximately 75%. These results suggest that the R-CNN method provides a more reliable and practical solution for classifying briquette quality, particularly in the presence of complex image backgrounds, highlighting the importance of advanced object detection techniques for real-world applications.
Norbertus Tri Suswanto Saptadi, P. Chyan, Jonatan Sentani· JOIV: International Journal...· 0 citations
This paper presents an object detection–based multimedia system for intelligent classification and grading of unmilled local rice using mobile imaging and deep learning. Unlike conventional image-level classification, the study reformulates rice grading as a multi-object detection problem, enabling grain-level localization and classification of clean, unclean, and wet grains within a single image. A YOLOv11-based model is integrated into a cloud-assisted architecture to support realtime inference in field conditions. The system was trained and evaluated on a balanced dataset of 1,350 annotated images collected from multiple rice varieties under realistic acquisition settings. Results demonstrate strong detection performance, achieving F1-scores between 0.97 and 0.99 across all classes. While F1-score is emphasized, the study acknowledges the need for additional metrics such as mean Average Precision (mAP) and confidence intervals for more comprehensive evaluation. YOLOv11 is selected for its efficiency in real-time deployment; however, future work will include benchmarking against models such as YOLOv8 and Faster R-CNN. The relatively modest dataset size is also recognized as a limitation. User evaluation using UTAUT indicates high usability and practical relevance. Overall, the system provides a scalable and transparent solution for preliminary rice grading and decision support.
Owen B. Pilongo, Bryle Kelly Dimabayaoa, Jimny Mackenzie H. Lu et al.· International Conference on...· 0 citations
This project aims to develop a deep learning model utilizing Convolutional Neural Networks (CNNs) to detect and categorize surface defects in metal photos. Traditional manual inspection is typically slow, uneven, and full of mistakes, with a research gap in automated solutions. The objective is to provide an effective and dependable solution for fault identification, crucial in manufacturing and quality control procedures. A proprietary CNN will be trained on a large dataset of labeled pictures, facilitating the extraction of critical characteristics via convolutional, pooling, and fully connected layers. The project will examine the usage of pre-trained models such as VGG16 and ResNet50. These models will be evaluated singly and in combination using an ensemble method to boost classification accuracy and resilience. The models’ performances will be measured based on evolution metrics such as accuracy, loss, and their ability to categorize photos into one of the defect categories. It is predicted that the ensemble strategy would give greater performance compared to individual models, illustrating the usefulness of deep learning-based methods for automated fault identification. The method will give a scalable deep learning framework for defect detection, enabling real time integration in manufacturing industries.
Prottoy Dev, Aninda Chowdhury, Mohd Abdul Haleem Abir· Engineering· 0 citations
The proposed framework provides reproducible reference results for evaluating supervised machine learning approaches to defective pixel detection while analyzing the influence of feature extraction window size on classification performance and reconstructed image quality.
Bárbaro M. López-Portilla, Kristian Balzer, Lorena Carballo et al.· Applied Sciences· 0 citations
