| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 7.05 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A doença da vespa-das-galhas-do-castanheiro (Dryocosmus kuriphilus) constitui uma das maiores ameaças à produção de castanha, afetando significativamente a saúde e produtividade dos castanheiros. Esta praga provoca deformações nas folhas e ramos dos castanheiros, resultando no enfraquecimento da árvore e numa diminuição considerável da produção de castanha. A identificação precoce e o combate eficiente à vespa-das-galhas são cruciais para minimizar os prejuízos e garantir a sustentabilidade da produção de castanha.
Este trabalho propõe um sistema de deteção automática da presença infeções nas galhas, utilizando técnicas de visão computacional, que visam facilitar a identificação precoce da praga e, assim, melhorar a eficácia das ações de controlo. O sistema foi desenvolvido com recurso a Convolutional Neural network (CNN) e ao modelo YOLOv8.2, treinado com um dataset composto por imagens anotadas de galhas de castanheiro. Este dataset foi construído e aumentado através de técnicas de data augmentation para melhorar a capacidade de generalização do modelo e aumentar a robustez do sistema perante diferentes condições de iluminação e ângulos de captura. A deteção das galhas infetadas é realizada através da análise de imagens capturadas por um drone, com uma velocidade controlada de forma a garantir a captura de detalhes suficientes para uma deteção precisa, ou outro meio de captação de imagens de alta definição. Esta abordagem permite realizar um rastreio eficiente e de baixo custo, cobrindo grandes áreas em menos tempo e com menor intervenção humana.
Os resultados obtidos demonstram que o sistema desenvolvido tem potencial para ser utilizado numa fase de pré-análise, oferecendo apoio à decisão para os produtores e cooperativas. Este sistema auxilia no combate à vespa-das-galhas-do-castanheiro, permitindo uma identificação precoce e uma intervenção direcionada, contribuindo para a redução dos danos causados pela praga. Apesar de ainda não estar totalmente maduro para aplicação final, o sistema mostra-se promissor na proteção das produções de castanha.
The chestnut gall wasp disease (Dryocosmus kuriphilus) is one of the greatest threats to chestnut production, significantly affecting the health and productivity of chestnut trees. This pest causes deformations in the leaves and branches of chestnut trees, resulting in tree weakening and a considerable reduction in chestnut yield. Early identification and efficient control of the gall wasp are crucial to minimize damage and ensure the sustainability of chestnut production. This work proposes an automatic detection system for the presence of infections in galls using computer vision techniques, aimed at facilitating the early identification of the pest and thus improving the effectiveness of control actions. The system was developed using Convolutional Neural Networks (CNN) and the YOLOv8.2 model, trained with a dataset composed of annotated images of chestnut galls. This dataset was constructed and augmented through data augmentation techniques to improve the model's generalization capability and increase the system's robustness under different lighting conditions and capture angles. The detection of infected galls is performed through the analysis of images captured by a drone, with a controlled speed to ensure the capture of sufficient details for accurate detection, or by other high-definition imaging means. This approach allows for efficient and cost-effective screening, covering large areas in less time and with minimal human intervention. The results obtained demonstrate that the developed system has the potential to be used in a pre-analysis phase, providing decision support to producers and cooperatives. This system assists in combating the chestnut gall wasp, enabling early identification and targeted intervention, contributing to reducing the damage caused by the pest. Although it is not yet fully mature for final application, the system shows promise in protecting chestnut production.
The chestnut gall wasp disease (Dryocosmus kuriphilus) is one of the greatest threats to chestnut production, significantly affecting the health and productivity of chestnut trees. This pest causes deformations in the leaves and branches of chestnut trees, resulting in tree weakening and a considerable reduction in chestnut yield. Early identification and efficient control of the gall wasp are crucial to minimize damage and ensure the sustainability of chestnut production. This work proposes an automatic detection system for the presence of infections in galls using computer vision techniques, aimed at facilitating the early identification of the pest and thus improving the effectiveness of control actions. The system was developed using Convolutional Neural Networks (CNN) and the YOLOv8.2 model, trained with a dataset composed of annotated images of chestnut galls. This dataset was constructed and augmented through data augmentation techniques to improve the model's generalization capability and increase the system's robustness under different lighting conditions and capture angles. The detection of infected galls is performed through the analysis of images captured by a drone, with a controlled speed to ensure the capture of sufficient details for accurate detection, or by other high-definition imaging means. This approach allows for efficient and cost-effective screening, covering large areas in less time and with minimal human intervention. The results obtained demonstrate that the developed system has the potential to be used in a pre-analysis phase, providing decision support to producers and cooperatives. This system assists in combating the chestnut gall wasp, enabling early identification and targeted intervention, contributing to reducing the damage caused by the pest. Although it is not yet fully mature for final application, the system shows promise in protecting chestnut production.
Description
Keywords
Vespa-das-galhas-do-castanheiro Deteção automática Visão computacional YOLOv8.2 Convolutional neural network Drone Rastreio de pragas