Complementary chemometrics and deep learning for semantic segmentation of tall and wide visible and near-infrared spectral images of plants

Mishra, Puneet; Sadeh, Roy; Bino, Ehud; Polder, G.; Boer, Martin P.; Rutledge, Douglas N.; Herrmann, Ittai

doi:10.1016/j.compag.2021.106226

Cited by 17 publications

(6 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although our approach showed excellent efficiency and precision in GT counting, we still found that it was challenging to identify and categorize overlapping GTs during female flowering, when a high density of GTs occurred. To further enhance the capabilities of CGTDM, a promising avenue for improvement lies in the integration of other spectral and image processing techniques, such as infrared imaging and hyperspectral imaging (Mishra et al., 2021; Wang, Shen, et al., 2021), which can enhance the feature extraction and discrimination capabilities in the context of cannabis plants. For algorithm optimization, the Segment Anything Model (Kirillov et al., 2023) for pixel‐level segmentation can contribute to the advancement of mask accuracy and segmentation efficiency.…”

Section: Discussionmentioning

confidence: 99%

Deep learning‐based quantification and transcriptomic profiling reveal a methyl jasmonate‐mediated glandular trichome formation pathway in Cannabis sativa

Huang,

Chen,

Zhao

et al. 2024

The Plant Journal

View full text Add to dashboard Cite

SUMMARYCannabis glandular trichomes (GTs) are economically and biotechnologically important structures that have a remarkable morphology and capacity to produce, store, and secrete diverse classes of secondary metabolites. However, our understanding of the developmental changes and the underlying molecular processes involved in cannabis GT development is limited. In this study, we developed Cannabis Glandular Trichome Detection Model (CGTDM), a deep learning‐based model capable of differentiating and quantifying three types of cannabis GTs with a high degree of efficiency and accuracy. By profiling at eight different time points, we captured dynamic changes in gene expression, phenotypes, and metabolic processes associated with GT development. By integrating weighted gene co‐expression network analysis with CGTDM measurements, we established correlations between phenotypic variations in GT traits and the global transcriptome profiles across the developmental gradient. Notably, we identified a module containing methyl jasmonate (MeJA)‐responsive genes that significantly correlated with stalked GT density and cannabinoid content during development, suggesting the existence of a MeJA‐mediated GT formation pathway. Our findings were further supported by the successful promotion of GT development in cannabis through exogenous MeJA treatment. Importantly, we have identified CsMYC4 as a key transcription factor that positively regulates GT formation via MeJA signaling in cannabis. These findings provide novel tools for GT detection and counting, as well as valuable information for understanding the molecular regulatory mechanism of GT formation, which has the potential to facilitate the molecular breeding, targeted engineering, informed harvest timing, and manipulation of cannabinoid production.

show abstract

Section: Discussionmentioning

confidence: 99%

Deep learning‐based quantification and transcriptomic profiling reveal a methyl jasmonate‐mediated glandular trichome formation pathway in Cannabis sativa

Huang,

Chen,

Zhao

et al. 2024

The Plant Journal

View full text Add to dashboard Cite

show abstract

“…In this paper, the attention mechanism was incorporated into the U-Net model to form the U-Net-Attention model, and it can be seen from Table 2 that the U-Net-Attention model had higher accuracy. Incorporating attention mechanisms into models to improve segmentation accuracy is a more common approach in deep learning, and in order to adapt to specific segmentation tasks, Mishra added the attention mechanism into the original model, which further improves the segmentation accuracy ( Mishra et al., 2021 ). In future work, we can try to add the attention module to different positions in the model to obtain better segmentation effects.…”

Section: Discussionmentioning

confidence: 99%

Non-destructive monitoring method for leaf area of Brassica napus based on image processing and deep learning

Liao

et al. 2023

Front. Plant Sci.

View full text Add to dashboard Cite

IntroductionLeaves are important organs for photosynthesis in plants, and the restriction of leaf growth is among the earliest visible effects under abiotic stress such as nutrient deficiency. Rapidly and accurately monitoring plant leaf area is of great importance in understanding plant growth status in modern agricultural production.MethodIn this paper, an image processing-based non-destructive monitoring device that includes an image acquisition device and image process deep learning net for acquiring Brassica napus (rapeseed) leaf area is proposed. A total of 1,080 rapeseed leaf image areas from five nutrient amendment treatments were continuously collected using the automatic leaf acquisition device and the commonly used area measurement methods (manual and stretching methods).ResultsThe average error rate of the manual method is 12.12%, the average error rate of the stretching method is 5.63%, and the average error rate of the splint method is 0.65%. The accuracy of the automatic leaf acquisition device was improved by 11.47% and 4.98% compared with the manual and stretching methods, respectively, and had the advantages of speed and automation. Experiments on the effects of the manual method, stretching method, and splinting method on the growth of rapeseed are conducted, and the growth rate of rapeseed leaves under the stretching method treatment is considerably greater than that of the normal treatment rapeseed.DiscussionThe growth rate of leaves under the splinting method treatment was less than that of the normal rapeseed treatment. The mean intersection over union (mIoU) of the UNet-Attention model reached 90%, and the splint method had higher prediction accuracy with little influence on rapeseed.

show abstract

“…21 Usually, the HSI preprocessing techniques for deep learning are similar to those of machine learning, 15 and deep learning can also benefit from conventional preprocessing methods. [22][23][24][25] There was also research that demonstrated that the appropriate CNN model could transform the input data to a suitable form for prediction and extract sufficient abstract features from raw spectrum so that the performance of quantitative analyzing tasks of spectroscopic data could be improved without data preprocessing. [26][27][28][29][30][31][32] The potential of an end-to-end analysis system based on raw spectral data without preprocessing is attractive because inappropriate preprocessing methods might remove useful information and computational resources could be saved by avoiding redundant optimization.…”

Section: Introductionmentioning

confidence: 99%

Optimizing Hyperspectral Imaging Classification Performance with CNN and Batch Normalization

Zhang,

Abdulla

2023

Applied Spectroscopy Practica

View full text Add to dashboard Cite

Background: Hyperspectral imaging systems face numerous challenges in acquiring accurate spatial-spectral hypercubes due to sample surface heterogeneity, environmental instability, and instrumental noise. Preprocessing strategies such as outlier detection, calibration, smoothing, and normalization are typically employed to address these issues, selecting appropriate techniques based on prediction performance evaluation. However, the risk of misusing inappropriate preprocessing methods remains a concern. Methods: In this study, we evaluate the impact of five normalization methods on the classification performance of six different classifiers using honey hyperspectral images. Our results show that different classifiers have varying compatible normalization techniques and that using Batch Normalization with Convolutional Neural Networks (CNN) can significantly improve classification performance and diminish the variations among other normalization techniques. The CNN with Batch Normalization can achieve a macro average F1 score of ≥0.99 with four different normalization methods and ≥0.97 without normalization. Furthermore, we analyze kernel weights' distribution in the CNN models' final convolutional layers using statistical measurements and kernel density estimation (KDE) graphs. Results: We find that the performance improvements resulting from adding BatchNorm layers are associated with kernel weight range, kurtosis, and density around 0. However, the differences among normalization methods do not show a strong correlation with kernel weight distribution. In conclusion, our findings demonstrate that the CNN with Batch Normalization layers can achieve better prediction results and avoid the risk of inappropriate normalization.

show abstract

Complementary chemometrics and deep learning for semantic segmentation of tall and wide visible and near-infrared spectral images of plants

Cited by 17 publications

References 40 publications

Deep learning‐based quantification and transcriptomic profiling reveal a methyl jasmonate‐mediated glandular trichome formation pathway in Cannabis sativa

Deep learning‐based quantification and transcriptomic profiling reveal a methyl jasmonate‐mediated glandular trichome formation pathway in Cannabis sativa

Non-destructive monitoring method for leaf area of Brassica napus based on image processing and deep learning

Optimizing Hyperspectral Imaging Classification Performance with CNN and Batch Normalization

Contact Info

Product

Resources

About