DDSC: Shallow Depthwise Separable Convolution Neural Network for Plant Leaf Diseases Classification

While most convolution neural network (CNN)-based approaches for plant disease classification improve accuracy through deeper architectures (stacked convolutional and pooling layers), this design paradigm inevitably incurs substantial computational overhead and parameter redundancy. The elevated parameter count particularly exacerbates overfitting risks when training data is limited. To tackle these issues, we introduce a novel shallow depthwise separable CNN (DDSC) for plant leaf diseases classification. The core of the DDSC lies in its use of depthwise separable convolution to achieve efficient feature extraction while significantly reducing both computational cost and parameter count compared to standard convolution. In addition, we employ <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1 \times 1$</tex> convolutions to enrich the feature representation. Experimental results demonstrate that DDSC achieves state-of-the-art classification accuracy while simultaneously maintaining superior computational efficiency, with significantly reduced parameter counts compared to baseline models.