<scp>EFFINCEP</scp> ‐ <scp>NET</scp> : A Proportional Inception‐Enhanced Modified <scp>EfficientNet</scp> ‐ <scp>B4</scp> Architecture for Rice Leaf Disease Recognition and Classification Incorporating Subtle Lesion Characteristics and Spatial Relati

ABSTRACT Recognising and categorising rice leaf diseases is crucial for sustaining high‐yielding and quality rice products. Leveraging advancements in computer vision, this research proposes an artificial intelligence‐powered hybrid convolutional network model, EFFINCEP‐NET, that combines a modified EfficientNet‐B4 architecture with an Inception module featuring a proportional filter relationship in each parallel convolutional operation, thereby improving rice leaf disease detection. This integration enhances the identification of subtle lesion characteristics and facilitates the comprehension of complex spatial relationships within infected leaf images. EFFINCEP‐NET achieves peak performance relatively quickly, reducing overall training time and computational resources. The dataset used to train the proposed model is a publicly accessible labelled collection of infected rice leaf images, comprising a total of 5932 samples, including four disease classes: Tungro, blast, bacterial blight and brown spot. Equated with existing baseline architectures such as DenseNet‐169 having accuracy of 97.68%, ResNet‐50 having accuracy of 98.01% and VGG‐19 having accuracy of 93.00%, the proposed EFFINCEP‐NET showed improved test accuracies of 99.43%, 98.65% and 95.60% across three different cross validation splits, with an average test classification accuracy of 98.65%, demonstrating its skill in capturing discriminative rice leaf disease with fewer parameters for classification in precision agriculture to improve crop health and yield.