A Hybrid Technique for Plant Disease Identification and Localisation in Real-time

TL;DR

The paper tackles real-time plant disease detection and localization in high-resolution field imagery, where traditional image processing and DNN methods struggle with computational load. It introduces a hybrid pipeline that combines a conditioned Quad-Tree recursive partitioning to localize ROI with a transfer-learned Xception classifier (Depthwise Separable Convolutions + residuals) for disease detection, followed by a BFS-based region grouping to output bounding boxes. On 10,000 images spanning Early/Late blight on potato/tomato from PlantVillage and field data, the method achieves an F1 around $0.80$ and demonstrates real-time inference on consumer GPUs, enabling potential drone/robot deployment. The work offers a practical framework for scalable crop-disease surveillance in precision agriculture, though it notes the need for more diverse real-world data and semi-supervised retraining to bridge laboratory and field conditions.

Abstract

Over the past decade, several image-processing methods and algorithms have been proposed for identifying plant diseases based on visual data. DNN (Deep Neural Networks) have recently become popular for this task. Both traditional image processing and DNN-based methods encounter significant performance issues in real-time detection owing to computational limitations and a broad spectrum of plant disease features. This article proposes a novel technique for identifying and localising plant disease based on the Quad-Tree decomposition of an image and feature learning simultaneously. The proposed algorithm significantly improves accuracy and faster convergence in high-resolution images with relatively low computational load. Hence it is ideal for deploying the algorithm in a standalone processor in a remotely operated image acquisition and disease detection system, ideally mounted on drones and robots working on large agricultural fields. The technique proposed in this article is hybrid as it exploits the advantages of traditional image processing methods and DNN-based models at different scales, resulting in faster inference. The F1 score is approximately 0.80 for four disease classes corresponding to potato and tomato crops.

Figures (9)

• Figure 1: Recursive Partitioning of image.
• Figure 2: An illustration to demonstrate the functioning of the conventional convolution process applied to a picture input with size 14 × 14 × 3, utilising a single kernel with dimensions 3x3x3
• Figure 3: Example of Depth-Wise Separable Convolutions (DWSC) operation using an image with dimensions of 14 by 14 by 3 as input. Three kernels of dimension 3x3x1 are used to perform a depth-wise convolution first, and then one kernel of dimension 1x1x3 is used to perform a pointwise convolution.
• Figure 4: Modified Xception model
• Figure 5: Flow chart of the algorithm