A method of corn disease recognition based on transfer learning using MobilenetV3 is proposed. The training dataset is augmented through online data enhancement, and the learning results of the MobilenetV3-Small network on the ImageNet dataset are utilized as pre-trained weights to construct the transfer learning model. A deep separable convolution module is adopted to reduce the model’s parameters count. Additionally, a channel attention mechanism and the H-Swish activation function are incorporated to enhance both the accuracy and efficiency of the model’s recognition capabilities. The Adam optimizer and cross-entropy loss function are used to train the top-level classifier after migration. Experimental results show that the model achieves an accuracy of 95.68% on the test set. Subsequently, the last one-third of layers in the transfer model are unfrozen, and the model is well tuned by adjusting the learning rate and optimizer parameters, resulting in a final test accuracy of 98.15%, which is an improvement of 2.47% compared to the pre-tuning accuracy.
KRIZHEVSKYA, SUTSKEVERI, HINTONG E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[2]
SIMONYANK, ZISSERMANA. Very deep convolutional networks for large-scale image recognition[EB/OL].(2014-09-04)[2023-03-14].
[3]
SZEGEDYC, LIUW, JIAY Q, et al. Going deeper with convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, MA, USA:IEEE, 2015: 1-9.
[4]
IOFFES, SZEGEDYC. Batch normalization: Accelerating deep network training by reducing internal covariate shift[EB/OL]. (2015-02-11)[2023-03-14].
[5]
SZEGEDYC, VANHOUCKEV, IOFFES, et al. Rethinking the inception architecture for computer vision[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA:IEEE, 2016: 2818-2826.
[6]
SZEGEDYC, IOFFES, VANHOUCKEV, et al. Inception-v4, inception-ResNet and the impact of residual connections on learning[J]. Proceedings of the AAAI Conference on Artificial Intelligence,2017,31(1):4278-4284.
[7]
HEKaiming, ZHANGXiangyu, RENShaoqing, et al. Deep residual learning for image recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA:IEEE, 2016: 770-778.
[8]
HUANGGao, LIUZhuang, VAN DER MAATENL, et al. Densely connected convolutional networks[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA:IEEE, 2017: 2261-2269.
CHOLLETF. Xception: Deep learning with depthwise separable convolutions[C]//IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA:IEEE, 2017: 1800-1807.
[19]
ZHANGXingyu, ZHOUXinyu, LINMengxiao, et al. ShuffleNet: An extremely efficient convolutional neural network for mobile devices[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City,UT, USA:IEEE,2018:6848-6856.
[20]
MANingning, ZHANGXiangyu, ZHENGHaitao, et al. ShuffleNet V2: Practical guidelines for efficient CNN architecture design[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer International Publishing, 2018: 116-131. DOI:10.1007/978-3-030- 01264-9_8 .
[21]
HOWARDA G, ZHUMenglong, CHENBo, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications[EB/OL].(2017-04-17)[2023-03-14].
[22]
HOWARDA, SANDLERM, CHENBo, et al. Searching for MobileNetV3[C]//IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Republic of Korea: IEEE, 2019: 1314-1324.