Maize Tassel Trait Characterization
Hao Lu, Zhiguo Cao, Yang Xiao, Zhiwen Fang, Yanjun Zhu, Ke Xian
Hao Lu, Zhiguo Cao, Yang Xiao, Zhiwen Fang, Yanjun Zhu, Ke Xian
In this work, we explore the feasibility of characterizing several traits of maize tassels with the computer vision technology. Technically, the trait characterization of maize tassels can be viewed as typical object detection and image segmentation problems from the perspective of computer vision. To address these problems, a fine-grained machine vision system termed mTASSEL is proposed. mTASSEL consists of four sub-modules designed for specific tasks. In addition, a novel maize tassel trait characterization dataset that contains 10 sequences and 16031 manually-annotated maize tassel instances is constructed by our team. We believe this dataset will be of great value for both computer vision and agriculture engineering communities. Both the dataset and source code of a complete implementation of the mTASSEL machine vision system can be downloaded in this page. More importantly, we hope one could realize that both tassel detection and segmentation in the field-based conditions are indeed challenging problems, which is beyond the ability of standard image processing techniques. There still needs efforts to work it out.
mTASSEL system involves four modules, i.e., mTASSEL-P, mTASSEL-D, mTASSEL-S, and mTASSEL-T. 'm' means 'multi-view'.
mTASSEL-P consists of a set of low-level image processing operations. The idea is to use color as a prior to extrat potential tassel regions.
mTASSEL-D plays a role that decides whether a potential belongs to the tassel or the background. To improve the robustness, multi-view visual representations (different feature views as well as multiple channel views) are utilized.
When a reliable tassel region is identified, mTASSEL-S further performs pixel-level semantic labeling to extract the fine-grained shape of tassel. We also consider multi-view representations.
At the final step, mTASSEL-T maps the segmentation result into some parameters with physical meanings, such as length, width, diameter, perimeter, color, and branch number.
The maize tassel trait characterization dataset is mainly developed for the evaluation of tassel detection, though the tassel segmentation subset is also provided. Each image in the detection subset is associated with a ".mat" file that can be loaded in Matlab. It specifies the ground-truth location of tassels. For more details of the dataset, please refer to our paper.
Maize Tassel Trait Characterization Dataset
Here we provide a highly vectorized Matlab implementation of the mTASSEL system as well as pre-trained models used to reproduce exactly the experimental results reported in our paper.
Source code and model can be downloaded from:
mTASSEL-Matlab-release-v1.0
Hao Lu, Zhiguo Cao, Yang Xiao, Zhiwen Fang, Yanjun Zhu, Ke Xian, Fine-grained maize tassel trait characterization with multi-view representations, Computers and Electronics in Agriculture, volume 118, pages 143-158, 2015.
@article{lu2015fine, title={Fine-grained maize tassel trait characterization with multi-view representations}, author={Lu, Hao and Cao, Zhiguo and Xiao, Yang and Fang, Zhiwen and Zhu, Yanjun and Xian, Ke}, journal={Computers and Electronics in Agriculture}, volume={118}, pages={143--158}, year={2015}, doi={10.1016/j.compag.2015.08.027},}