Intelligent detection of citrus fruit pests using machine vision system and convolutional neural network through...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Plant pests, such as the Mediterranean fruit fly, cause substantial economic losses in agriculture annually.
• Accurate and early pest detection is crucial for effective crop management and minimizing yield reduction.

Purpose of the Study:

• To investigate the efficacy of machine vision and deep learning for early detection of Mediterranean fruit fly infestation in citrus fruits.
• To compare the performance of four pre-trained Convolutional Neural Network (CNN) models (ResNet-50, GoogleNet, VGG-16, AlexNet) using different optimization algorithms.

Main Methods:

• Collected 1519 citrus fruit images under natural light conditions across three infestation stages.
• Utilized 70% of images for training, 10% for validation, and 20% for testing deep learning models.
• Employed ResNet-50, GoogleNet, VGG-16, and AlexNet CNN architectures with SGDm, RMSProp, and Adam optimizers for image classification.

Main Results:

• VGG-16 with SGDm achieved 98.33% accuracy and 98.36% F1-score in the early pest outbreak stage.
• AlexNet with SGDm demonstrated superior performance in the later stage with 99.33% accuracy and 99.34% F1-score.
• AlexNet with SGDm exhibited the fastest network training time at 323 seconds.

Conclusions:

• Machine vision systems combined with deep learning, particularly CNNs like AlexNet and VGG-16, are highly effective for early pest detection in agricultural products.
• These technologies offer a promising non-destructive approach for real-time pest monitoring and management in orchards.