# ApaltAI: a web-based diagnostic system with a sequential voting architecture for detecting anthracnose and scab in avocado fruit

## TL;DR

ApaltAI is a web-based system that uses deep learning to accurately detect anthracnose and scab in avocado fruits, helping reduce crop losses.

## Contribution

A novel binary sequential voting architecture (VotingBS) is introduced for automated avocado disease detection with high accuracy.

## Key findings

- The system achieved 98.92% precision, 98.89% recall, and 99.03% accuracy in disease detection.
- A two-stage deep learning ensemble was used to classify healthy vs. diseased fruits and identify specific diseases.
- The web application includes modules for crop management and phytosanitary analysis, aiding farmers and technicians.

## Abstract

Avocado (Persea americana Mill.), with a global production estimated at 10.4 million tons in 2023, suffers annual losses of 20-30% due to diseases such as anthracnose (Colletotrichum gloeosporioides) and scab (Sphaceloma perseae), resulting in substantial economic impacts for major producing countries (Mexico, Peru, and Colombia). This study introduces an advanced system that integrates a binary sequential voting architecture (VotingBS) with a fully functional web application, for the automated identification of two high-incidence diseases: anthracnose and scab, both of which critically affect fruit quality and yield. The proposed VotingBS architecture implements a hierarchical two-stage classification strategy. In the first stage, a five-model deep learning ensemble differentiates between healthy and diseased fruits. In the second stage, another ensemble determines which of the two diseases is present. For this purpose, a collection of 674 labeled fruit images was used for training and validation. Experimental results demonstrate outstanding model performance, achieving key metrics such as 98.92% precision, 98.89% recall, and 99.03% accuracy, significantly outperforming traditional approaches. Moreover, the solution was deployed through a web app featuring dedicated modules for crop management, phytosanitary analysis, and disease diagnosis. This architecture enhances the system’s practical utility and facilitates its adoption by farmers, field technicians, and agricultural monitoring agencies. Overall, this work demonstrates how combining hybrid deep learning models with accessible digital platforms can revolutionize plant disease diagnostics, fostering a more efficient, automated, and resilient precision agriculture.