Se-Mobilenetv2-Based Nine-Class Insect Image Recognition
DOI:
https://doi.org/10.54097/xpzvd972Keywords:
Computer Vision; Deep Learning; Image Processing; Smart Farming.Abstract
Accurate and automated identification of crop pests is vital for modern agriculture, yet remains challenging due to the visual similarity of species, environmental variability, and computational constraints. This paper proposes a SE-MobileNetV2 network, which integrates Squeeze-and-Excitation (SE) modules into MobileNetV2, enhancing channel-wise feature recalibration for robust insect classification. The approach leverages advanced data augmentation, label smoothing, and cosine annealing learning rate scheduling to address data imbalance and improve generalisation. Evaluated on a balanced nine-class pest image dataset from Kaggle, SE-MobileNetV2 achieves 100% validation accuracy by the sixth epoch. It demonstrates superior inference speed and efficiency compared to ResNet baselines, making it suitable for mobile and edge deployment. Experimental results highlight the model's advantages in accuracy, parameter efficiency, and deployability, while error analysis identifies areas for further improvement. The findings provide a solid technical foundation for intelligent, practical pest recognition systems in agriculture, with future directions including transfer learning and multimodal data integration.
Downloads
References
[1] Howard A G, Zhu M, Chen B, et al. MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861, 2017.
[2] Sandler M, Howard A, Zhu M, et al. MobileNetV2: Inverted residuals and linear bottlenecks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018: 4510–4520.
[3] Hu J, Shen L, Sun G. Squeeze-and-Excitation Networks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018: 7132–7141.
[4] He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770–778.
[5] Zhang X, Zhou X, Lin M, Sun J. ShuffleNet: An extremely efficient convolutional neural network for mobile devices. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018: 6848–6856.
[6] Ma N, Zhang X, Zheng H T, Sun J. ShuffleNet V2: Practical guidelines for efficient CNN architecture design. Proceedings of the European Conference on Computer Vision (ECCV), 2018: 116–131.
[7] Fuentes A, Yoon S, Kim S C, et al. A robust deep-learning-based detector for real-time tomato plant diseases and pests’ recognition. Sensors, 2017, 17(9): 2022.
[8] Loshchilov I, Hutter F. Decoupled weight decay regularisation. International Conference on Learning Representations (ICLR), 2019.
[9] Müller R, Kornblith S, Hinton G. When does label smoothing help. Advances in Neural Information Processing Systems (NeurIPS), 2019, 32: 4696–4705.
[10] Liu W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector. Proceedings of the European Conference on Computer Vision (ECCV), 2016: 21–37.
[11] Liu R, Cui L, Wang J, et al. Research on pest identification based on deep learning and knowledge graph. Computers and Electronics in Agriculture, 2020, 178: 105796.
[12] Redmon J, Farhadi A. YOLOv3: An incremental improvement. arXiv preprint arXiv:1804.02767, 2018.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Highlights in Science, Engineering and Technology
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
