Cayco, Francis Mark N. Marquez, Joshua Marc D. Mercullio, Miraiza Elisa V. Paclian, Micha Lene R. Villar, Helen Claire J. 4 0
Integratedn mosquito detection and fumigation system utilizing faster region-based convolutional neural networks. 6 6 - - - xviii, 256 pp. 28 cm. - - - - - . - . - 0 . - . - 0 .
Undergraduate Thesis: (Bachelor of Science in Computer Engineering) - Pamantasan ng Lungsod ng Maynila, 2024.
5
ABSTRACT: STATEMENT OF THE PROBLEM: Mosquitos, carriers of diseases like malaria, dengue fever, Zika virus, and West Nile virus, pose a global public health threat, particularly in endemic regions. These diseases can lead to serve illness, death, and economic burdens. Conventional mosquito control methods often use pesticides, which have limitations in precision and can harm non-target species while fostering pesticide resistance in mosquitos. To address these challenges, a more precise and targeted approach is required. Urbanization is increasing the health threat by mosquito-borne diseases. This growth creates more potential breeding sites for mosquitos. The current mosquito detection and fumigation systems suffer from a critical lack of integration and comprehensive analytics. While some fumigation systems are in place, they operate independently without offering insights into the extent of mosquito infestation or the effectiveness of fumigation efforts. RESEARCH METHODOLOGY: This research adopts a quantitative approach, integrating computer science, internet of things, and web development to create an automated mosquito control system. The system features a multi-domain architecture for image-based mosquito detection and targeted fumigation. It includes: 1. Detection Domain - Processes images using an ESP-32 Camera and uploads the photo to the backend server for machine learning processing. Also contains the luring mechanism designed to attract mosquitos to the detection platform. 2. Fumigation Domain - A subsystem controlled by the ESP-32 Wi-Fi controlled by the spraying system. Automated fumigation is triggered via a 12-V Relay Module. 3. Machine Learning Domain - Contains the Faster RCNN Model where datasets are obtained and augmented; and model is trained and tested. 4. Website Domain - A publicly-available website that allows stakeholders to view mosquito-related data. User interaction is facilitated through a web interface. SUMMARY FINDINGS: The detection system's neural network showed significant improvement during the training phase, with metrics like Region Proposal Regression and Class Regression Loss enhancing region and bounding box accuracy. The model demonstrated high precision and recall for larger objects, through ir struggled with smaller ones, and required fine-tuning for crowded scences. The luring system, tested with acetic acid, UV light, and heat pads, revealed acetic acid as a superior lure, effectively drawing mosquitos compared to dry ice. UV light's effectiveness varied with environmental conditions, while heat pads consistently attracted mosquitos by simulating body heat. Rge automated fumigation system, controlled by a microcontroller and ESP32 module, efficiently used a 12-V Relay Module and DC micro diaphragm pump to spray pesticide, with each 5L tank lasting up to 18 fumigations and reaching 1.5-2 meters. The real-time data analytics website dashboard performed well, meeting metrics for load time, responsiveness, and data visualization. It integrated multiple detection and fumigation systems for comprehensive data analysis but faced hardware limitations preventing direct system interaction. User satisfaction survey showed that respondents had a positive impression, believing the system could mitigate negative effects and reduce disease spread. CONCLUSION: The interdisciplinary approach adopted in this study, combining computer science, machine learning, and web development, holds promise for transforming mosquito control efforts in the barangay community of Tondo, Manila, and beyond. By leveraging technology to mitigate the threat of mosquito-borne diseases, we move closer to a future where communities are safer, healthier, and more resilient in the face of emerging public health challenges. RECOMMENDATIONS: This project demonstrated the potential of a multi-domain architecture for automated mosquito control, effectively using image capture, machine learning-based detection and counting, and targeted fumigation. To optimize the systems's efficacy, user experience, and environmental impact, several advancements are recommended. First, leveraging faster GPU's to reduce training time and exploring alternative neural network models like YOLOv4 enhance performance. For luring mechanisms, testing different attractant chemicals and rigorously evaluating equipment effectiveness beyond integration testing are crucial. In terms of fumigation, using less pervasive chemicals for safety, expanding spray coverage, and increasing chemical container capacity will improve operational duration and safety. Enhancing the website by reducing latency, integrating faster APIs, and adding visual aids like graphs and charts will improve the user experience. To optimize energy efficiency and scalability, minimizing wires and cords, expanding system coverage, and incorporating solar panels and batteries for a more independent and sustainable prototype are essential.
5
2 = =
2
2 --0------
6 --0-- 2 --------
0 2 --
--20------
--------20--
--------20--
----2
/ 2
/ 2
/
/
Integratedn mosquito detection and fumigation system utilizing faster region-based convolutional neural networks. 6 6 - - - xviii, 256 pp. 28 cm. - - - - - . - . - 0 . - . - 0 .
Undergraduate Thesis: (Bachelor of Science in Computer Engineering) - Pamantasan ng Lungsod ng Maynila, 2024.
5
ABSTRACT: STATEMENT OF THE PROBLEM: Mosquitos, carriers of diseases like malaria, dengue fever, Zika virus, and West Nile virus, pose a global public health threat, particularly in endemic regions. These diseases can lead to serve illness, death, and economic burdens. Conventional mosquito control methods often use pesticides, which have limitations in precision and can harm non-target species while fostering pesticide resistance in mosquitos. To address these challenges, a more precise and targeted approach is required. Urbanization is increasing the health threat by mosquito-borne diseases. This growth creates more potential breeding sites for mosquitos. The current mosquito detection and fumigation systems suffer from a critical lack of integration and comprehensive analytics. While some fumigation systems are in place, they operate independently without offering insights into the extent of mosquito infestation or the effectiveness of fumigation efforts. RESEARCH METHODOLOGY: This research adopts a quantitative approach, integrating computer science, internet of things, and web development to create an automated mosquito control system. The system features a multi-domain architecture for image-based mosquito detection and targeted fumigation. It includes: 1. Detection Domain - Processes images using an ESP-32 Camera and uploads the photo to the backend server for machine learning processing. Also contains the luring mechanism designed to attract mosquitos to the detection platform. 2. Fumigation Domain - A subsystem controlled by the ESP-32 Wi-Fi controlled by the spraying system. Automated fumigation is triggered via a 12-V Relay Module. 3. Machine Learning Domain - Contains the Faster RCNN Model where datasets are obtained and augmented; and model is trained and tested. 4. Website Domain - A publicly-available website that allows stakeholders to view mosquito-related data. User interaction is facilitated through a web interface. SUMMARY FINDINGS: The detection system's neural network showed significant improvement during the training phase, with metrics like Region Proposal Regression and Class Regression Loss enhancing region and bounding box accuracy. The model demonstrated high precision and recall for larger objects, through ir struggled with smaller ones, and required fine-tuning for crowded scences. The luring system, tested with acetic acid, UV light, and heat pads, revealed acetic acid as a superior lure, effectively drawing mosquitos compared to dry ice. UV light's effectiveness varied with environmental conditions, while heat pads consistently attracted mosquitos by simulating body heat. Rge automated fumigation system, controlled by a microcontroller and ESP32 module, efficiently used a 12-V Relay Module and DC micro diaphragm pump to spray pesticide, with each 5L tank lasting up to 18 fumigations and reaching 1.5-2 meters. The real-time data analytics website dashboard performed well, meeting metrics for load time, responsiveness, and data visualization. It integrated multiple detection and fumigation systems for comprehensive data analysis but faced hardware limitations preventing direct system interaction. User satisfaction survey showed that respondents had a positive impression, believing the system could mitigate negative effects and reduce disease spread. CONCLUSION: The interdisciplinary approach adopted in this study, combining computer science, machine learning, and web development, holds promise for transforming mosquito control efforts in the barangay community of Tondo, Manila, and beyond. By leveraging technology to mitigate the threat of mosquito-borne diseases, we move closer to a future where communities are safer, healthier, and more resilient in the face of emerging public health challenges. RECOMMENDATIONS: This project demonstrated the potential of a multi-domain architecture for automated mosquito control, effectively using image capture, machine learning-based detection and counting, and targeted fumigation. To optimize the systems's efficacy, user experience, and environmental impact, several advancements are recommended. First, leveraging faster GPU's to reduce training time and exploring alternative neural network models like YOLOv4 enhance performance. For luring mechanisms, testing different attractant chemicals and rigorously evaluating equipment effectiveness beyond integration testing are crucial. In terms of fumigation, using less pervasive chemicals for safety, expanding spray coverage, and increasing chemical container capacity will improve operational duration and safety. Enhancing the website by reducing latency, integrating faster APIs, and adding visual aids like graphs and charts will improve the user experience. To optimize energy efficiency and scalability, minimizing wires and cords, expanding system coverage, and incorporating solar panels and batteries for a more independent and sustainable prototype are essential.
5
2 = =
2
2 --0------
6 --0-- 2 --------
0 2 --
--20------
--------20--
--------20--
----2
/ 2
/ 2
/
/