With the increase of biological images, how to classify them effectively is a challenging problem, the Convolutional Neural Networks (CNNs) show promise for this problem. The challenges of using CNNs to handle images classification lie in two aspects: (1) How to further improve the classification accuracy? (2) How to make the network more light weight? To address the above challenges, this paper proposed a biological image classification method based on improved CNN. In this paper, fixed size images as input of CNN are replaced with appropriately large size images and some modules were replaced with an Inverted Residual Block module with fewer computational cost and parameters. The proposed method extensively evaluated the computational cost and classification accuracy on five well known benchmark datasets, and the results demonstrate that compared with existing image classification methods, proposed method shows better performance image classification and reduces the network parameters and computational cost.

随着生物图像的增加，如何有效地对其进行分类是一个具有挑战性的问题，卷积神经网络（CNN）对此问题显示了希望。使用CNN进行图像分类的挑战主要体现在两个方面：（1）如何进一步提高分类精度？（2）如何使网络重量更轻？针对上述挑战，本文提出了一种基于改进的CNN的生物图像分类方法。在本文中，将固定大小的图像作为CNN的输入替换为适当大小的图像，并且将某些模块替换为具有较少计算成本和参数的反向残差块模块。拟议的方法广泛评估了五个众所周知的基准数据集的计算成本和分类准确性，