Analyn Niere Yumang

Efficient inventory monitoring remains a critical challenge for retailers, particularly in beverage refrigerators where products are depth-stacked and often occluded. Traditional manual audits and object detection methods are labor-intensive, error-prone, and limited to visible items, resulting in inaccurate stock records and poor replenishment decisions. This study proposes a real-time refill notification system by integrating YOLOv8 and MiDaS monocular depth estimation. Using a single front-facing camera connected to a Raspberry Pi 4, the system detects beverage products, estimates depth to infer hidden stock, and provides color-coded feedback—Green for full stock, Yellow for moderate stock, and Orange for low stock. Performance evaluation with a multiclass confusion matrix demonstrates the system’s ability to monitor visible and occluded items under varying stocking conditions accurately. Results show improvements in inventory visibility, reduced reliance on manual audits, and timely restocking alerts, making the solution particularly suitable for small to medium-sized retailers. Beyond refrigerators, the framework may be extended to other retail and vending environments, providing a scalable approach to AIdriven shelf monitoring.

Accurate size classification remains a persistent challenge for agricultural products with irregular morphology, such as bitter melon (Momordica charantia). Proper grading is essential for fair pricing, efficient packaging, and compliance with the Association of Southeast Asian Nations and Philippine National Standards, yet traditional manual sorting often results in inconsistencies. To address this, we introduce an automated classification framework built on the You Only Look Once Version 8 (YOLOv8) model. The system integrates Harris Corner Detection to enhance feature extraction and the Douglas–Peucker algorithm to simplify contour representations, thereby reducing noise and improving shape analysis. A dataset of Ampalaya images was trained and processed to detect and categorize fruit sizes, with evaluation conducted through a confusion matrix. Experimental results showed an overall classification accuracy of 93.75%, demonstrating that the combined approach effectively balances precision with computational efficiency. Beyond improving classification accuracy, the findings highlight the broader potential of combining deep learning and contour-based methods to advance agricultural automation, optimize post-harvest workflows, and strengthen competitiveness in both local and international markets.

A YOLOv8-based instance segmentation model is developed for detecting and segmenting surface corrosion in steel hand tools in this study. The model was trained using a 7000-image custom dataset and deployed on a Raspberry Pi 5-based setup with a controlled lighting environment. The testing results showed an overall segmentation precision of 86.62%, a recall of 82.90%, and an F1-score 84.72%. Corrosion segmentation struggles with a precision of 75.02%, a recall of 17.40%, an F1-score of 28.25% and a Dice coefficient of 72.21%, demonstrating effective tool detection and classification, but it struggles on small corrosion patches, emphasizing the need for architectural and dataset enhancements.

We developed an augmented Sungka board that integrates traditional Filipino gameplay with embedded sensor technology. Each pit is equipped with load cell sensors and HX711 analog-to-digital converters to accurately detect marble distribution and movement in real time. A Raspberry Pi 4 serves as the central controller, handling sensor data acquisition, game state processing, rule enforcement, and output display through a liquid crystal display. The system enables automatic score tracking, move validation, and real-time board updates without altering the physical structure or rules of Sungka. A rule-based decision algorithm using fuzzy logic and heuristic search evaluates possible moves in constant time, allowing seamless real-time interaction. Across 10,000 simulated games, the algorithm achieved win rates of 84.9% against random, 77.7% against greedy, and 56.3% against exact-match strategies, with statistically consistent performance. By combining reliable hardware sensing with intelligent decision support, the proposed system enhances engagement while preserving the cultural authenticity of Sungka.

Guava is a popular fruit in the Philippines, as it offers various health benefits. Its leaves are used in traditional medicine to aid in wound healing, stomach disorders, pain relief, and more. In this study, we classified guava leaf diseases using Support Vector Machine (SVM) and You Only Look Once version 8 (YOLOv8). Raspberry Pi 4 is used to control the image preprocessing and the program that utilizes the proposed trained model. The SVM model conducts image classification, while YOLOv8 handles feature extraction and object detection. Grayscale and color thresholding segmentation feature extraction is also implemented in the proposed model. The developed model combines both YOLOv8 and SVM algorithms to evaluate their accuracy using a confusion matrix, achieving a 92.5% accuracy. With its very low error rate, the system can accurately classify guava leaf diseases.

Facial features are critical visual indicators for understanding what a person is experiencing, providing valuable insights into their emotions and physical states. However, accurately detecting these features under diverse conditions remains a significant challenge, especially in computationally constrained environments. This paper presents a facial feature extraction method designed to identify regions of interest for detecting facial cues, with a focus on improving the accuracy of eye and mouth detection. Addressing the limitations of standard Haar cascade classifiers, particularly in challenging scenarios such as droopy eyes, red eyes, and droopy mouths, this method introduces a correction algorithm rooted in normal human facial anatomy, emphasizing symmetry and consistent feature placement. By integrating this correction algorithm with a feature-based refinement process, the proposed approach enhances detection accuracy from 67.22% to 96.11%. Through this method, the accurate detection of facial features like the eyes and mouth is significantly improved, offering a lightweight and efficient solution for real-time applications while maintaining computational efficiency.

Accurately identifying road conditions, particularly wetness, is crucial for ensuring road safety and enhancing vehicle performance. We conducted road surface classification and road wetness estimation using state-of-the-art deep learning models in this study. Raspberry Pi Model 4 was used to classify road surfaces and estimate road wetness. SqueezeNet, a lightweight convolutional neural network, was used to recognize wet and dry road surfaces with an accuracy of 90%. The ENet model, known for its efficiency in semantic segmentation tasks, was used to estimate the degree of wetness, categorizing roads into damp, wet, and very wet roads. The ENet model showed an accuracy of 90.48%. The efficiency of the deep learning models in road surface wetness monitoring was validated using a confusion matrix created with the margin classifier. A total of 300 images per category were used for training, amounting to 1200 in total. A total of 20 testing images were used for road surface classification and 21 for road wetness estimation. The results highlighted the robustness and applicability of SqueezeNet and ENet models in estimating diverse environmental road conditions.

Exploring different activation functions can help in understanding their impact on the model's performance, stability, and convergence speed. It can also lead to discovering the most suitable activation function for specific tasks or datasets. This study investigated various activation functions for YOLOv8, specifically ReLU, Mish, Hardswish, PReLU, and FReLU, in comparison to the default activation function, SiLU. The CRDDC2022 dataset was employed to evaluate and compare the performance metrics of each activation function. The findings revealed that Mish and Hardswish performed similarly, achieving mean average precision scores of 0.449 and 0.439, respectively, indicating their potential as promising alternatives to SiLU. Conversely, PReLU and FReLU did not yield satisfactory results, highlighting challenges that warrant further research in this area.

In recent studies, many researchers have tested different kind of techniques focusing on flowers, barks and leaf for plant recognition such as extraction of shape factors and comparison of color histogram. Both have provided good results but has many restrictions. This study on leaf shape recognition using Fourier descriptors and Hough transform with probabilistic neural network has shown great potential as an improvement for recognition of plant species. It is a combination between shape factors, Fourier descriptors, and vein features using Hough transform. All image processing techniques are done using MATLAB. The hardware implementation emulated wireless connection between a thin client and a server with the help of Mono, OpenCV, and EmguCV. Hardware peripherals are composed of a miniature studio, Raspberry Pi processor as well as its Pi Cam and LCD modules. The mini studio was used to provide clean background and correct lighting. Results show an accuracy of 87 percent on Flavia dataset, and 99 percent on the local dataset.