Remote fruit fly detection using computer vision and machine learning-based electronic trap

• Published in: Frontiers in plant science 2023-10, Vol.14, p.1241576-1241576
• Main Authors: Molina-Rotger, Miguel, Morán, Alejandro, Miranda, Miguel Angel, Alorda-Ladaria, Bartomeu
• Format: Article
• Language: English
• Subjects: computer vision; machine learning; olive fruit fly pest; Plant Science; precision agriculture; random forest; support vector machine
• ISSN: 1664-462X; 1664-462X
• DOI: 10.3389/fpls.2023.1241576

IntroductionIntelligent monitoring systems must be put in place to practice precision agriculture. In this context, computer vision and artificial intelligence techniques can be applied to monitor and prevent pests, such as that of the olive fly. These techniques are a tool to discover patterns and abnormalities in the data, which helps the early detection of pests and the prompt administration of corrective measures. However, there are significant challenges due to the lack of data to apply state of the art Deep Learning techniques.MethodsThis article examines the detection and classification of the olive fly using the Random Forest and Support Vector Machine algorithms, as well as their application in an electronic trap version based on a Raspberry Pi B+ board.ResultsThe combination of the two methods is suggested to increase the accuracy of the classification results while working with a small training data set. Combining both techniques for olive fly detection yields an accuracy of 89.1%, which increases to 94.5% for SVM and 91.9% for RF when comparing all fly species to other insects.DiscussionThis research results reports a successful implementation of ML in an electronic trap system for olive fly detection, providing valuable insights and benefits. The opportunities of using small IoT devices for image classification opens new possibilities, emphasizing the significance of ML in optimizing resource usage and enhancing privacy protection. As the system grows by increasing the number of electronic traps, more data will be available. Therefore, it holds the potential to further enhance accuracy by learning from multiple trap systems, making it a promising tool for effective and sustainable fly population management.