Search papers, labs, and topics across Lattice.
This paper investigates the use of Histogram Matching (HM) to improve the robustness of ResNet-18 models for grapevine disease detection under variable illumination conditions. HM is applied both as a preprocessing normalization step and as a data augmentation technique to diversify the training data. Results show significant improvements in classification accuracy on real-world canopy images, particularly when HM is used for both normalization and augmentation, suggesting it effectively addresses domain gaps caused by lighting variations.
Simple histogram adjustments can dramatically improve the real-world robustness of plant disease classifiers, especially in uncontrolled lighting conditions.
Variability in illumination is a primary factor limiting deep learning robustness for field-based plant disease detection. This study evaluates Histogram Matching (HM), a technique that transforms the pixel intensity distribution of an image to match a reference profile, to mitigate this in grapevine classification, distinguishing among healthy leaves, downy mildew, and spider mite damage. We propose a dual-stage integration of HM: (i) as a preprocessing step for normalization, and (ii) as a data augmentation technique to introduce controlled training variability. Experiments using 1,469 RGB images (comprising homogeneous leaf-focused and heterogeneous canopy samples) to train ResNet-18 models demonstrate that this combination significantly enhances robustness on real-world canopy images. While leaf-focused samples showed marginal gains, the canopy subset improved markedly, indicating that balancing normalization with histogram-based diversification effectively bridges the domain gap caused by uncontrolled lighting.
