Plant Leaf Classification Based on Efficient Attention Mechanism ECSA

In this study, we proposed an Efficient Channel and Spatial Attention module (ECSA), which aims to improve the performance of convolutional neural networks in plant leaf image classification tasks by combining spatial and channel attention. The ECSA module can make full use of attention information while keeping the model lightweight, and enhance the model's recognition ability of fine grain characteristics of plant leaves. Experiments on the self-made small plant leaf dataset and the public dataset Cifar-10 prove that ECSA module has better performance than other mainstream attention modules in image classification tasks. In addition, the ablation experiment further analyzed the effects of channel attention and spatial attention on the model performance in different dimensions, as well as the effects of serial and parallel information acquisition methods. It was found that channel attention had the most significant improvement on the model performance, and the serial information acquisition method was slightly better than the parallel method. At the same time, the influence of cross-information interaction region size on the model performance was also discussed, and the optimal region size on the plant leaf dataset was determined. Overall, the ECSA module shows remarkable potential in processing fine-grained plant leaf image classification problems, and has good generalization ability.