Deep Learning-Based Computational Model for Disease Identification in Cocoa Pods (Theobroma cacao L.)
Darlyn Buenaño Vera, Byron Oviedo, Washington Chiriboga Casanova, Cristian Zambrano-Vega
TL;DR
This work tackles early disease identification in cocoa pods (monilia and black pod) by building a deep learning–based object detector using EfficientDet-Lite4, trained on a curated image dataset and deployed as an Android app (Cacao DL) for on-field use. A seven-architecture comparison guides the selection of EfficientDet-Lite4 due to its balance of localization accuracy, speed, and mobile viability, achieving a representative mean average precision of about 0.30 at $IoU=0.5$ and approximately 0.23 at $IoU$ thresholds up to 0.95, with per-class precision differences (e.g., ~42.3% for fitoftora, ~27.3% for monilia, ~34.4% for healthy). The dataset combines Mocache plantation images (three classes) with Kaggle validation data, and images are cropped to 640×640 RGB and labeled for training; training occurs on Google Colab with TF-Lite Model Maker and is exported to TensorFlow Lite for on-device inference. The on-device mobile app demonstrates a practical tool enabling farmers to diagnose pod health and access treatment guidance, signaling important implications for disease management in cocoa production, while future work aims to scale up data, test additional architectures, and assess performance across devices.
Abstract
The early identification of diseases in cocoa pods is an important task to guarantee the production of high-quality cocoa. The use of artificial intelligence techniques such as machine learning, computer vision and deep learning are promising solutions to help identify and classify diseases in cocoa pods. In this paper we introduce the development and evaluation of a deep learning computational model applied to the identification of diseases in cocoa pods, focusing on "monilia" and "black pod" diseases. An exhaustive review of state-of-the-art of computational models was carried out, based on scientific articles related to the identification of plant diseases using computer vision and deep learning techniques. As a result of the search, EfficientDet-Lite4, an efficient and lightweight model for object detection, was selected. A dataset, including images of both healthy and diseased cocoa pods, has been utilized to train the model to detect and pinpoint disease manifestations with considerable accuracy. Significant enhancements in the model training and evaluation demonstrate the capability of recognizing and classifying diseases through image analysis. Furthermore, the functionalities of the model were integrated into an Android native mobile with an user-friendly interface, allowing to younger or inexperienced farmers a fast and accuracy identification of health status of cocoa pods