Lightweight Pyramidal Multi-Scale Attention Residual Network for Plant Disease Recognition

Earlier plant disease recognition plays a crucial role in determining agricultural yield. Therefore, developing an effective method for accurately identifying plant disease categories has become a significant concern within the agricultural domain. In this article, we propose a lightweight model for plant disease recognition called Lightweight Pyramidal Multi-scale Attention Residual Network (LPMA-ResNet). Our model incorporates a dual pyramidal structure to capture comprehensive information and utilizes an adaptive selection attention mechanism to identify informative features from pyramid multi-scale features while suppressing irrelevant information. To evaluate the performance of our model, extensive experiments were conducted on the soybean diseases dataset and plant village dataset. The experimental results obtained from these plant diseases datasets demonstrate that our model outperforms other state-of-the-art methods by achieving superior performance with fewer parameters utilized. LIMA-ResNet (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times 1.0$ </tex-math></inline-formula>) could achieve over 2.12% accuracy improvement compared to ShuffleNet-v2(<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times 2.0$ </tex-math></inline-formula>) while using only about 2/3 of its parameters on the soybean dataset. Meanwhile, we provide visualizations illustrating the impact of our proposed method on enhancing model performance to validate its effectiveness. Based on these empirical experiments, we demonstrate the effectiveness of our approach to plant disease recognition.