Introduction

We address the problem of maize tassl flowering status recognition by Computer Vision. Three categories of non-flowering, partially-flowering, and fully-flowering are considered. By leveraging the annotations of our previous work, a flowering status dataset with 3000 images is also constructed.

As maize plants grow from the tasseling stage to flowering stage (two critical growth stages of maize), we obeserve that their tassels exhibit three types of status described above. To our knowledge, it is the first time for the intermediate partially-flowering status to be considered.

Monitoring the flowering status of maize tassel accurately has great significance in refining the farming operation for production enhancement. However, existing solutions to monitor the flowering status still excessively depend on human observations. This cannot meet the requriment of large-scale observations in the real-world field environment. Actually, the efficiency of observation can largely benefit from partly automating this process. With the help of modern Computer Vision and Machine Learning technologies, we attempt to explore the feasibility to recognize automatically the maize tassel flowering status in an image-based paradigm.

Technical Pipeline

We follow a standard image categorization pipeline to extract visual representations, i.e., densely sampled SIFT following by Fisher Vector encoding. However, we consider such representations are not good enough for a fine-grained task. To boost the accuracy further, a novel large-margin dimensionality reduction (LMDR) formulation is proposed. LMDR not only enhances the discriminative power of representations but also reduces the memory comsumption.

Dataset

Code

The MATALB implementation of our approach is also made available here. It can be accessed via the links below:

MTFS-Matlab-release-v1.0

• Google Drive
• Baidu Yun

Publication

Hao Lu, Zhiguo Cao, Yang Xiao, Zhiwen Fang, Yanjun Zhu, Towards fine-grained maize tassel flowering status recognition: Dataset, theory, and practice, Applied Soft Computing, volume 56, pages 34-45, 2017

Citation

@article{lu2017mtfs,

  title={Towards fine-grained maize tassel flowering status recognition: dataset, theory and practice},

  author={Lu, Hao and Cao, Zhiguo and Xiao, Yang and Fang, Zhiwen and Zhu, Yanjun},

  journal={Applied Soft Computing},

  volume={56},

  pages={34--45},

  year={2017},

  doi={10.1016/j.asoc.2017.02.026}

}