A Comparative Evaluation of Image Pre-processing Techniques for Corn Leaf Disease Classification

The swift progression of deep learning has markedly advanced corn leaf disease detection, enabling highly efficient and accurate classification from corn leaf imagery. Yet, the influence of image pre-processing on model performance remains a critical and nuanced area of inquiry, particularly for complex, real-world agricultural datasets. This study offers a comparative assessment of prominent pre-processing techniques Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gaussian blur on the performance of several leading deep convolutional neural network architectures, including VGG16, ResNet50, InceptionV3, Xception, MobileNetV2, DenseNet121, and EfficientNetB0, for corn leaf disease classification. Utilizing a publicly available corn leaf disease dataset from Kaggle, we systematically applied and examined the impact of each pre-processing method, both individually and in combination, across all selected models. Notably, our findings indicate that the highest classification accuracy 98% was achieved using the ResNet50 architecture with Gaussian blur as the sole pre-processing step. This result surpasses the performance of the same model under other pre-processing conditions and exceeds that of all other model-pre-processing combinations evaluated. While CLAHE and the combined use of both pre-processing techniques yielded moderate gains for specific architectures, neither approach surpassed the ResNet50 with Gaussian blur. These outcomes suggest that judicious application of certain image enhancement methods can still deliver meaningful improvements, even as advanced networks demonstrate robust capability to compensate for raw image variability internally. This work highlights the necessity of critically benchmarking pre-processing strategies, rather than adopting them by default, and provides valuable guidance for researchers and practitioners in constructing resilient disease classification pipelines for precision agriculture.