Web Application of Convolutional Neural Networks with YOLOv8 for Early Detection of Diseases in Strawberry Crops

In this research, we address a critical problem for strawberry growers in Lima: the management of phytosanitary diseases. Traditionally, these farmers relied on time-consuming, imprecise, and error-prone visual observation methods, resulting in annual production losses of 49.44%. We developed a comprehensive technological system based on convolutional neural networks (CNNs) using the YOLOv8 architecture, specifically designed to identify diseases such as powdery mildew, anthracnose, and gray mold, representing a significant shift toward precision agriculture methodologies. Our research was applied, with a quasi-experimental design and a quantitative approach. We worked with 474 high-resolution images of strawberry crops from 38 producers in Manchay Alto, Pachacamac district. Statistical analysis using SPSS version 27 with the Wilcoxon signed-rank test revealed statistically significant results (p = 0.000), achieving a very good technical accuracy of 96.74% (mAP@50) and remarkable system effectiveness, with 84.4% of cases reaching a high level. The system demonstrated superior performance compared to traditional inspection methods, facilitating timely disease detection and accurate diagnoses. Agronomic validation by local experts confirmed 91- 94% accuracy for four locally present diseases, while identifying systematic false positives for three diseases not present under Lima’s specific microclimatic conditions, revealing a critical gap between international training datasets and local disease prevalence that has significant implications for agricultural AI deployment in diverse agroclimatic regions.