Exploration of identifying apple leaf diseases using lightweight convolutional neural network model
Although the increase of network depth and width can enhance the recognition accuracy,it is often not suitable for mobile device applications due to the large number of parameters and calculations required.To address this problem,we developed two lightweight CNN models.These models improve the feature extraction ability of the network by enhancing the Fire module of the SqueezeNet network,incorporating a spatial attention mechanism,and introducing a dense connection module in the deep layer within the network.Through training on an apple disease leaf dataset,the recognition accuracy of the improved model reaches 89.60％and 94.37％,exceeding the original network's accuracy by 2.98％and 7.75％,respectively.Remarkably,the number of parameters in these networks remains low,at only 0.9M and 2.5M.The experimental results showed that the improved network not only maintains a lightweight model but also achieves higher recognition accuracy.
NETL
NSTL
万方数据
