| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.45 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Cerca de um terƧo da produĆ§Ć£o global de alimentos depende da polinizaĆ§Ć£o das abelhas, tornando-as vitais para a economia mundial. No entanto, existem diversas ameaƧas Ć sobrevivĆŖncia das espĆ©cies de abelhas, tais como incĆŖndios florestais, stress humano induzido, subnutriĆ§Ć£o, poluiĆ§Ć£o, perda de biodiversidade, agricultura intensiva e predadores como as vespas asiĆ”ticas. Destes problemas, pode-se observar um aumento da necessidade de soluƧƵes automatizadas que possam auxiliar na monitorizaĆ§Ć£o remota de colmeias de abelhas. O objetivo desta tese Ć© desenvolver soluƧƵes baseadas em Aprendizagem AutomĆ”tica para problemas que podem ser identificados na apicultura, usando tĆ©cnicas e conceitos de Deep Learning, VisĆ£o Computacional e Processamento de Sinal. Este documento descreve o trabalho da tese de mestrado, motivado pelo problema acima exposto, incluindo a revisĆ£o de literatura, anĆ”lise de valor, design, planeamento de testes e validaĆ§Ć£o e o desenvolvimento e estudo computacional das soluƧƵes. Concretamente, o trabalho desta tese de mestrado consistiu no desenvolvimento de soluƧƵes para trĆŖs problemas ā classificaĆ§Ć£o da saĆŗde de abelhas a partir de imagens e a partir de Ć”udio, e deteĆ§Ć£o de abelhas e vespas asiĆ”ticas. Os resultados obtidos para a classificaĆ§Ć£o da saĆŗde das abelhas a partir de imagens foram significativamente satisfatĆ³rios, excedendo os que foram obtidos pela metodologia definida no trabalho base utilizado para a tarefa, que foi encontrado durante a revisĆ£o da literatura. No caso da classificaĆ§Ć£o da saĆŗde das abelhas a partir de Ć”udio e da deteĆ§Ć£o de abelhas e vespas asiĆ”ticas, os resultados obtidos foram modestos e demonstram potencial aplicabilidade das respetivas metodologias desenvolvidas nos problemas-alvo. Pretende-se que as partes interessadas desta tese consigam obter informaƧƵes, metodologias e perceƧƵes adequadas sobre o desenvolvimento de soluƧƵes de IA que possam ser integradas num sistema de monitorizaĆ§Ć£o da saĆŗde de abelhas, incluindo custos e desafios inerentes Ć implementaĆ§Ć£o das soluƧƵes. O trabalho futuro desta dissertaĆ§Ć£o de mestrado consiste em melhorar os resultados dos modelos de classificaĆ§Ć£o da saĆŗde das abelhas a partir de Ć”udio e de deteĆ§Ć£o de objetos, incluindo a publicaĆ§Ć£o de artigos para obter validaĆ§Ć£o pela comunidade cientĆfica.
Up to one third of the global food production depends on the pollination of honey bees, making them vital for the world economy. However, between forest fires, human-induced stress, poor nutrition, pollution, biodiversity loss, intensive agriculture, and predators such as Asian Hornets, there are plenty of threats to the honey bee speciesā survival. From these problems, a rise of the need for automated solutions that can aid with remote monitoring of bee hives can be observed. The goal of this thesis is to develop Machine Learning based solutions to problems that can be identified in beekeeping and apiculture, using Deep Learning, Computer Vision and Signal Processing techniques and concepts. The current document describes master thesisā work, motivated from the above problem statement, including the literature review, value analysis, design, testing and validation planning and the development and computational study of the solutions. Specifically, this master thesisā work consisted in developing solutions to three problems ā bee health classification through images and audio, and bee and Asian wasp detection. Results obtained for the bee health classification through images were significantly satisfactory, exceeding those reported by the baseline work found during literature review. On the case of bee health classification through audio and bee and Asian wasp detection, these obtained results were modest and showcase potential applicability of the respective developed methodologies in the target problems. It is expected that stakeholders of this thesis obtain adequate information, methodologies and insights into the development of AI solutions that can be integrated in a bee health monitoring system, including inherent costs and challenges that arise with the implementation of the solutions. Future work of this master thesis consists in improving the results of the bee health classification through audio and the object detection models, including publishing of papers to seek validation by the scientific community.
Up to one third of the global food production depends on the pollination of honey bees, making them vital for the world economy. However, between forest fires, human-induced stress, poor nutrition, pollution, biodiversity loss, intensive agriculture, and predators such as Asian Hornets, there are plenty of threats to the honey bee speciesā survival. From these problems, a rise of the need for automated solutions that can aid with remote monitoring of bee hives can be observed. The goal of this thesis is to develop Machine Learning based solutions to problems that can be identified in beekeeping and apiculture, using Deep Learning, Computer Vision and Signal Processing techniques and concepts. The current document describes master thesisā work, motivated from the above problem statement, including the literature review, value analysis, design, testing and validation planning and the development and computational study of the solutions. Specifically, this master thesisā work consisted in developing solutions to three problems ā bee health classification through images and audio, and bee and Asian wasp detection. Results obtained for the bee health classification through images were significantly satisfactory, exceeding those reported by the baseline work found during literature review. On the case of bee health classification through audio and bee and Asian wasp detection, these obtained results were modest and showcase potential applicability of the respective developed methodologies in the target problems. It is expected that stakeholders of this thesis obtain adequate information, methodologies and insights into the development of AI solutions that can be integrated in a bee health monitoring system, including inherent costs and challenges that arise with the implementation of the solutions. Future work of this master thesis consists in improving the results of the bee health classification through audio and the object detection models, including publishing of papers to seek validation by the scientific community.
Description
Keywords
VisĆ£o Computacional Deep Learning MonitorizaĆ§Ć£o da SaĆŗde de Abelhas Processamento de Sinal Computer Vision Deep Learning Bee Health Monitoring Signal Processing