Browsing by Author "Marzan, Charlie S."

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Automated tobacco grading using image processing techniques and a convolutional neural network
    (International Association of Computer Science and Information Technology, 2019) Marzan, Charlie S.; Ruiz, Conrado R.
    Tobacco grading is very important for crop market price determination. It is beneficial for graders who need to manually classify tobacco leaves according to their grades. As such, a grading system using image processing techniques and a Convolutional Neural Network (CNN) is proposed in this study which aims to surpass existing results in tobacco grading. The system consists of image acquisition, pre-processing, leaf detection, segmentation, and classification. Tobacco leaf images were directly taken at the tobacco grading room and pre-processed for subsequent tasks. Through a Haar cascade classifier and applying image processing techniques, air-cured tobacco leaves are automatically detected and extracted in images. This method produced satisfactory results as it can successfully detect single and multiple tobacco leaves taken under different positions and scale. All detected tobacco leaves underwent various image processing to precisely segment leaves from the rest of the image. The experimental results also reveal that using segmented and nonsegmented images, CNN classifier can effectively grade tobacco leaves as high as 96.25% accuracy rate and on average, took 7.43 ms to classify a single tobacco leaf. This approach outperforms current methods in grading tobacco leaves. © 2019 International Association of Computer Science and Information Technology.
  • Thumbnail Image
    Item
    Deep learning-assisted milkfish fingerling counting:
    (Association for Computing Machinery, 2024) Sibayan, Christian Jose; Abando, Benjie R.; Acosta, Justin P.; Madrelijos, Johnico D.; Marzan, Charlie S.
    The manual counting of milkfish fingerlings within the fish industry in the Philippines poses significant challenges due to its time-consuming nature, labor intensiveness, and susceptibility to errors. To address these issues, this study presents a deep learning approach for the real-time counting of milkfish fingerlings to enhance both the speed and accuracy of fingerling counts. The dataset was meticulously curated from six milkfish fingerlings farms, with gathered videos converted into image frames and subjected to polygon annotation. Furthermore, various augmentation techniques were applied to these images to enhance training efficacy before integration into a deep neural network. Notably, implementing the YOLOv8n-seg object detector with segmentation and counting yielded a remarkable accuracy rate of up to 90.64% and 93.34%, respectively. This innovative methodology promises significant advancements in the automation and efficiency of milkfish fingerling counting processes within the fish industry, paving the way for improved productivity and accuracy in fishery management practices. © 2024 ACM.
  • Thumbnail Image
    Item
    Multi-task deep learning for automated tobacco leaf grading in a controlled environment
    (International Journal of Computer Theory and Engineering, 2026) Marzan, Charlie S.; Ruiz, Conrado Jr.; Aran, Oya
    Grading tobacco leaves is crucial for ensuring fair pricing and quality control, however, the process is still largely carried out manually, resulting in a slow, subjective, and often inconsistent outcome. In this work, we present a multi-task deep learning approach designed to automate the grading of air-cured Burley tobacco leaves in controlled settings. The model is constructed with shared convolutional layers and separate task-specific branches, allowing it to predict stalk group, quality, and color at the same time, in line with the hierarchical grading system. To improve consistency, images were preprocessed using coin-based size normalization, rotation alignment, and segmentation. In our experiments, the multi-task model with EfficientNetB0 achieved an accuracy of 94.82% and significantly outperformed the multi-class and single-task baselines, while reducing both training time and inference delay. These findings suggest that multi-task learning can be a valuable and robust method for automated tobacco grading, showing gains in accuracy, speed, and scalability compared to other algorithms.
  • Thumbnail Image
    Item
    Sign to speech convolutional neural network-based Filipino sign language hand gesture recognition system
    (2021 International Symposium on Computer Science and Intelligent Controls (ISCSIC), 2021) Jarabese, Mark Benedict D.; Marzan, Charlie S.; Boado, Jenelyn Q.; Lopez, Rushaine Rica Mae F.; Ofiana, Lady Grace B.; Pilarca, Kenneth John P.
    Sign Language Recognition is a breakthrough for helping deaf-mute people and has been studied for many years. Unfortunately, every research has its own limitation and are still unable to be used commercially. In this study, we developed a real-time Filipino sign language hand gesture recognition system based on Convolutional Neural Network. A manually gathered dataset consists of 237 video clips with 20 different gestures. This dataset underwent data cleaning and augmentation using image pre-processing techniques. The Inflated 3D convolutional neural network was used to train the Filipino sign language recognition model. The experiments considered retraining the pretrained model with top layers and all layers. As a result, the model retrained with all layers using imbalanced dataset was shown to be more effective and achieving accuracy up to 95% over the model retrained with top layers to classify different signs or hand gestures. Using the Rapid Application Development model, the Filipino sign language recognition application was developed and assessed its usability by the target users. With different parameters used in the evaluation, the application found to be effective and efficient.
  • Thumbnail Image
    Item
    Towards tobacco leaf detection using Haar cascade classifier and image processing techniques
    (Association for Computing Machinery, 2018-10) Marzan, Charlie S.; Marcos, Nelson
    Tobacco grading needs an effective leaf detection algorithm to ensure accurate results in segmentation and feature extraction. Leaf detection in this research used Haar cascade classifier and image processing techniques to automatically detect tobacco leaves in images. The proposed detection algorithm was implemented through OpenCV Python. The Haar cascade classifier was trained with 1,000 images and tested with 150 images. To improve the detection results of the classifier and ultimately detecting tobacco leaves, image processing techniques such as converting RGB to grayscale, blurring, thresholding, and finding connected components were applied. The experimental results show that the classifier can successfully distinguish tobacco leaves from other objects even those having resemblance to the characteristics of tobacco leaves in terms of color and shape. The accuracy rate of at least 91.33% proves the capability of the Haar cascade classifier to detect single and multiple tobacco leaves posed at different angles and taken at different distances from the camera. After applying some image processing techniques, the detection rate reached 100.00% and took 62 ms on average.