ResNet-50-Based Vision System for the Classification of Tetranychus urticae in Hass Avocado

Abstract

This paper presents the design of a computer vision system based on convolutional neural networks (CNNs) for the detection of Tetranychus urticae in Hass avocado leaves. This pest represents a significant phytosanitary challenge affecting the export quality of Peruvian avocados. The proposed approach uses the ResNet-50 architecture with transfer learning to classify images into two categories: healthy and infested leaves. The model was trained with a proprietary dataset of 900 images, collected under controlled lighting conditions, and split into 70% for training, 15% for validation, and 15% for testing. The images were preprocessed using normalization, resizing to 224×224 pixels, and data augmentation techniques to improve model robustness. The resulting system achieved a classification accuracy of 96% on the test dataset, confirming its effectiveness under controlled conditions. Although the system demonstrated high accuracy in a simulated environment, further validation with field data is necessary to assess its generalization and deployment feasibility in real agricultural scenarios. The development process adhered to the VDI 2206 methodology for mechatronic systems design and was implemented using Python, TensorFlow, Keras, and Jupyter Notebook. This study constitutes a validated proof of concept for intelligent pest classification in precision agriculture, with future work focusing on system validation in field conditions, integration with mobile platforms, and expansion to detect multiple pest types.