Crop diseases pose a major threat to global food security by reducing crop yields and increasing production losses. Okra is susceptible, particularly during the rainy season. The primary objective of this research is to deploy the Enhanced ResNeXt-50 model for prompt and accurate recognition of okra ( Abelmoschus esculentus ) disease severity levels, which is essential for efficient intervention and resource sustainability. Traditional manual inspection techniques are time-intensive, subjective, and not scalable. Although recent advancements in deep learning have shown promise, these methods often struggle with limited feature extraction, poor regularisation, vanishing gradient issues due to non-smooth activation functions, and high computational complexity, which hinder their deployment on low-power farming devices. To address these limitations, we propose an Enhanced ResNeXt-50 model for okra leaf disease severity level classification that incorporates several key innovations, including the Swish activation function to enable smoother backpropagation and mitigate vanishing gradients, a Stochastic pooling layer to minimise overfitting and improve regularisation, and Cat Swarm Optimisation (CSO) for feature selection and hyperparameter tuning. A Cardinality value of 48 is employed to improve multi-path feature representation. The model is trained and tested using our real-time dataset “Okra DiseaseNet” dataset comprising 2,500 images of healthy and diseased okra leaves with six severity levels classification. Our proposed enhanced ResNeXt-50 model achieved a superior accuracy of 96.01%, outperforming state-of-the-art models. This research demonstrates that the Enhanced ResNeXt-50 model provides a reliable and robust technique for severity classification of okra leaf diseases.
K. et al. (Fri,) studied this question.