Deep Learning Approaches for Cocoa Pod Disease Classification A Literature Review

Abstract

Cocoa (Theobroma cacao) is a cornerstone of many tropical economies, yet its yield is persistently threatened by pod diseases such as black pod rot, frosty pod rot, and cocoa pod borer infestation. Over the past decade, deep learning, and convolutional neural networks (CNNs) in particular, has emerged as a powerful tool for automated plant disease diagnosis from images. This paper presents a structured literature review of deep-learning approaches applied, directly or by close analogy, to cocoa pod disease classification. Following a PRISMA style protocol, 41 studies published between 2016 and 2025 were selected from major databases and synthesized along five dimensions: data sources and dataset construction, preprocessing and augmentation, network architectures, training and transfer-learning strategies, and evaluation methodology. The review finds that transfer learning with compact architectures, notably ResNet, MobileNet, and EfficientNet variants, dominates recent work and consistently achieves reported accuracies above 90% on related tasks. Three persistent gaps are identified: the scarcity of large, balanced, and openly available cocoa specific image datasets; limited validation under realistic field conditions; and inconsistent reporting of evaluation metrics. The review concludes by outlining research directions, including domain adaptation, lightweight on device inference, explainability, and standardized benchmarking, to move cocoa pod disease classification from controlled experiments toward deployable tools for smallholder agriculture.