Journal of IKEEE (전기전자학회논문지)

Institute of Korean Electrical and Electronics Engineers (한국전기전자학회)
권호 표지
DOI QR코드

DOI QR Code

A new Mada-CenterNet based on Dual Block to improve accuracy of pest counting

해충 카운팅의 정확성 향상을 위한 Dual Block 기반의 새로운 Mada-CenterNet

  • Hee-Jin Gwak (Dept. of Computer Engineering, Andong National University) ;
  • Cheol-Hee Lee (Dept. of Computer Engineering, Andong National University) ;
  • Chang-Hwan Son (Dept. of Software Science & Engineering, Kunsan National University)
  • 곽희진 ;
  • 이철희 ;
  • 손창환
  • Received : 2024.08.23
  • Accepted : 2024.09.24
  • Published : 2024.09.30
Download PDF
⟨ Previous Next ⟩

Abstract

Effective pest control in the agricultural field is essential for improving crop productivity. To do so, information on the type and timing of pests, as well as the amount of pests generated, is required. Mada-CenterNet, a prior study on pest counting, which is a method of identifying the amount of pest occurrence, has improved the accuracy of pest counting by utilizing transformable convolution and multiscale attention fusion and is reported to be the best in the field. In this study, a new transformer structure with a dual block was applied instead of multiscale attention, which is the transformer structure of Mada-CenterNet. More sophisticated feature maps were extracted through cross-attention of pixel path and semantic path. As a result of the experiment, the proposed model has improved the accuracy of pest counting. It is better than the existing Mada-CenterNet and effectively alleviates obstruction problems, damage to pests' bodies, and detection difficulties caused by various appearances. Unlike conventional pest counting methods, it can secure the advantage of reducing manpower and time costs, and it is expected that it can be used in other agricultural fields that require counting of objects.

농업 분야에서 해충에 대한 효과적인 방제는 작물 생산성 향상에 필수적인 요소이다. 해충의 방제를 위해서는 해충의 종류, 발생 시기는 물론이며, 해충의 발생량에 대한 정보가 필요하다. 해충의 발생량을 파악하는 방법인 해충 카운팅 관련 선행 연구인, Mada-CenterNet은 변형 가능한 컨볼루션과 멀티스케일 어텐션 퓨전을 활용하여 해충 카운팅의 정확도를 향상시켰으며 해당 분야에서 가장 우수하다고 보고되고 있다. 본 연구에서는 Mada-CenterNet의 트랜스포머 구조인 멀티스케일 어텐션 대체하는 새로운 트랜스포머 구조인 듀얼 블록을 적용하였으며, 픽셀 경로와 시맨틱 경로의 교차 어텐션을 통해 더욱 정교한 특징 맵을 추출하였다. 실험 결과, 제안된 모델은 기존 Mada-CenterNet보다 해충 카운팅 정확도가 우수함과 동시에 폐색 문제와 해충의 몸체 손상, 다양한 모습으로 인한 탐지의 어려움을 효과적으로 완화하였다. 기존 해충 카운팅의 방법과 달리, 인력 및 시간 비용을 절감할 수 있다는 장점을 확보할 수 있으며 물체의 계수가 필요한 다른 농업 분야에도 활용이 가능할 것으로 기대된다.

Keywords

Acknowledgement

This work was carried out with the support of the Cooperative Research Program for Agriculture Science & Technology Development (grant no: PJ016303) and the National Institute of Crop Science (NICS), Rural Development Administration (RDA), Republic of Korea.

References

  1. K. Yeo, I. Yu, H. Rhee, G. Choi, S. Lee and J. Lee, "Effects of ventilation systems and set point temperature of single-span plastic greenhouse on disease incidence, fruit quality and yield of oriental melon," In Korean Journal of Agricultural Science, vol.42, no.4, pp.325-333, 2015. DOI: 0.7744/cnujas.2015.42.4.325 https://doi.org/10.7744/cnujas.2015.42.4.325
  2. S. Kim and J. Ahn, "A YOLO-Based Crop Pests Detection Mobile Application for Smart Farming," Journal of the Korea Academia-Industrial cooperation Society, vol.25, no.7, pp.603-610, 2024. DOI: 10.5762/KAIS.2024.25.7.603
  3. Y. Jeong, S. Kim and D. Kim, "Implementation of Detection and Classification System for Sudden Pest Using Object Detection Algorithm," Journal of Korean Institute of Communications and Information Sciences, vol.48, no.6, pp.704-711, 2023. DOI: 10.7840/kics.2023.48.6.704
  4. K. Jo, S. Jung and C. Sim. "A Study of Shiitake Disease and Pest Image Analysis based on Deep Learning," Journal of Korea Multimedia Society, vol.23, no.1, pp.50-57, 2020. DOI: 10.9717/kmms.2020.23.1.050
  5. Joseph Redmon, Santosh Divvala, Ross Girshick and Ali Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.779-788, 2016. DOI: 10.48550/arXiv.1506.02640
  6. Shaoqing Ren, Kaiming He, Ross Girshick and Jian Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," IEEE transactions on pattern analysis and machine intelligence, vol.39, no.6, pp.1137-1149, 2016. DOI: 10.48550/arXiv.1506.01497
  7. Lin, Tsung-Yi, et al. "Focal loss for dense object detection," Proceedings of the IEEE international conference on computer vision, pp.2980-2988, 2017. DOI: 10.1109/iccv.2017.324
  8. Yang, Ze, et al. "Reppoints: Point set representation for object detection," Proceedings of the IEEE/CVF international conference on computer vision, pp.9657-9666, 2019. DOI: 10.48550/arXiv.1904.11490
  9. Duan, Kaiwen, et al. "Centernet: Keypoint triplets for object detection," Proceedings of the IEEE/CVF international conference on computer vision. pp.6569-6578, 2019. DOI: 10.48550/arXiv.1904.08189
  10. Zhang, Yingying, et al. "Single-image crowd counting via multi-column convolutional neural network," Proceedings of the IEEE conference on computer vision and pattern recognition. pp.589-597, 2016. DOI: 10.48550/arXiv.1603.06995
  11. L. Yuhong, X. Zhang, and D. Chen, "Csrnet: Dilated convolutional neural networks for understanding the highly congested scenes," Proceedings of the IEEE conference on computer vision and pattern recognition, pp.1091-1100, 2018. DOI: 10.48550/arXiv.1802.10062
  12. W. Jia, Q. Wang, and A. Chan, "Kernel-based density map generation for dense object counting." IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.44, no.3. pp.1357-1370, 2020. DOI: 10.1109/TPAMI.2020.3022878
  13. Ranjan, Viresh, H. Le, and M. Hoai, "Iterative crowd counting." Proceedings of the European conference on computer vision (ECCV), pp.270-285, 2018. DOI: 10.48550/arXiv.1807.09959
  14. J. Lee and C. Son. "Trap-based Pest counting: Multiscale and deformable attention centerNet integrating internal lr and hr joint feature learning," Remote Sensing, vol.15, no.15, 2023. DOI: 10.3390/rs15153810
  15. Dai Jifeng, et al. "Deformable convolutional networks," Proceedings of the IEEE international conference on computer vision, pp.765-773, 2017. DOI: 10.48550/arXiv.1703.06211
  16. Yao, Ting, et al. "Dual vision transformer," IEEE transactions on pattern analysis and machine intelligence, vol.45, no.9, pp.10870-10882, 2023. DOI: 10.48550/arXiv.2207.04976
  17. Vaswani, Ashish. "Attention is all you need," In Proceedings of the Advances in Neural Information Processing Systems, pp.5998-6008, 2017. DOI: 10.48550/arXiv.1706.03762