Advances in UAV-based deep learning for cassava disease monitoring and detection: A comprehensive review of models, imaging techniques, and agricultural applications

Abstract

ABSTRACT Cassava is a vital staple crop in sub-Saharan Africa, yet its production is severely threatened by diseases such as Cassava Mosaic Disease (CMD) and Cassava Brown Streak Disease (CBSD), which contribute to significant yield and economic losses. Traditional manual inspection methods for disease identification are time-consuming and unreliable, especially over large or remote farms. Recent advances in precision agriculture, particularly the integration of unmanned aerial vehicles (UAVs) and deep learning, have introduced scalable and accurate alternatives. This review explores two distinct approaches for cassava disease detection: Transformer-based model tailored for UAV-acquired imagery and a transfer learning method using the Inception v3 convolutional neural network with static leaf datasets. The Transformer-based ED-Swin architecture incorporates multi-scale attention and deformable convolution modules, enabling it to handle complex field conditions such as occlusion, uneven lighting, and irregular lesion morphology. On the other hand, the Inception v3 model, though computationally simpler, achieves competitive accuracy using relatively small datasets, making it ideal for smartphone-based field deployment. While the ED-Swin model excels in spatial awareness and classification accuracy, it demands higher computational resources. However, the Inception v3 approach offers practical advantages in low-resource environments due to its efficiency and ease of implementation. The review highlights how these complementary approaches can jointly inform a robust cassava disease monitoring framework, combining the high precision of UAV-Transformer systems with the accessibility of mobile Convolutional neural networks tools. Together, they underscore the transformative role of artificial intelligence in enhancing crop health surveillance and supporting sustainable agricultural practices across diverse farming environments

Topics & Concepts