Revolutionizing wheat yield prediction: Introducing SPSI for enhanced panicle number estimation using UAV imagery
image: (A to C) Coefficient of determination (R2 ) values for linear relationships of PNPA with SIs over individual dates before heading for different datasets. The values of the valueless regions were lower than 0.00.
Credit: Plant Phenomics
Wheat is crucial for global food security, and panicle number per unit ground area (PNPA) is key to its yield. Traditional manual counting methods are accurate but inefficient, prompting a shift towards remote sensing and image processing for rapid, nondestructive PNPA estimation. Recent studies have primarily used near-ground platforms for accurate, small-scale PNPA estimates, but their efficiency is limited. Unmanned aerial vehicles (UAVs) offer a promising alternative, especially pre-heading, using multispectral imagery to effectively manage and increase yields. However, challenges remain, such as spectral saturation affecting accuracy and the need for improved methods integrating spectral and textural analysis to overcome this. Additionally, the impact of exposed background materials on these estimates is not fully understood, and further research is needed to refine PNPA estimation techniques for wheat.
In September 2023, Plant Phenomics published a research article titled “SPSI: A Novel Composite Index for Estimating Panicle Number in Winter Wheat before Heading from UAV Multispectral Imagery ”.
This study introduced a spectral-textural panicle number per unit ground area (PNPA) sensitive index (SPSI) derived from unmanned aerial vehicle (UAV) multispectral imagery to improve PNPA estimation in winter wheat before heading by mitigating spectral saturation. The SPSI combined an optimal spectral index (SI) and textural index (TI) to address the effects of background materials on PNPA estimates. The performance of SPSI was compared with traditional SIs and TIs, revealing that green-pixel TIs generally outperformed all-pixel TIs, with specific exceptions. SPSI demonstrated superior overall accuracies and significantly reduced spectral saturation compared to other indices. Moreover, it showed improvements in correlation coefficients and reductions in root mean square error and relative root mean square error when applied to two experimental datasets.
The relationships between PNPA and various indices were examined, revealing that specific SIs exhibited stronger relationships with PNPA. The texture-based indices derived from green pixels exhibited significant differences in performance, with green-pixel-based TIs generally providing higher correlation coefficients. The research identified sensitive bands for constructing SPSI, noting the consistent and overlapping bands across different dates. This led to the identification of two COR-based normalized difference texture indices (NDTICORs) that were particularly sensitive to PNPA before booting. SPSI, combining DATT[850,730,675] and NDTICOR[850,730], was closely associated to PNPA, inheriting the advantages of both indices and demonstrating less sensitivity to cultivation factors.
The study further investigated the sensitivity of SPSI to various spectral uniformity scenarios and cultivation factors, indicating its robustness and lower sensitivity compared to DATT[850,730,675]. Modeling and validation results consistently affirmed the superior performance of SPSI's in PNPA estimation across different datasets, particularly around the optimal timing of estimation. In conclusion, the research concluded that incorporating textural information into a composite index effectively mitigates spectral saturation and improves PNPA estimation, offering potential benefits for crop yield prediction and precision agriculture, especially when applied to high-resolution satellite imagery.
###
References
Authors
Yapeng Wu1, Wenhui Wang1,2, Yangyang Gu1, Hengbiao Zheng1, Xia Yao1, Yan Zhu1, Weixing Cao1, and Tao Cheng1*
Affiliations
1National Engineering and Technology Center for Information Agriculture (NETCIA), MARA Key Laboratory for Crop System Analysis and Decision Making, MOE Engineering Research Center of Smart Agriculture, Jiangsu Key Laboratory for Information Agriculture, Nanjing Agricultural University, One Weigang, Nanjing, Jiangsu 210095, PR China.
2Langfang Normal University, 100 Aimin West Road, Langfang, Hebei 065000, PR China.
About Tao Cheng
Since December 2013, he has been a Professor with the National Engineering and Technology Center for Information Agriculture (NETCIA) and the College of Agriculture, Nanjing Agricultural University, Nanjing, China. His main research interests include crop mapping and monitoring, crop phenotyping, imaging and non-imaging spectroscopy of vegetation, quantitative remote sensing, machine learning, and image analysis.