Innovative deep learning model enhances maize fen. newswise

Newswise – A research team developed Point-Line Net, a deep learning method based on the Mask R-CNN framework, to automatically recognize corn field images and determine the number and growth trajectory of leaves and stalks. The model achieved an object detection accuracy (MAP50) of 81.5% and introduced a new lightweight keypoint detection branch. This innovative method promises to increase the efficiency of plant breeding and phenotype detection in complex field environments, paving the way for more accurate crop management and yield prediction.

Maize is an important crop globally, essential for food, feed and industrial applications. Understanding maize phenotypes, such as plant height, number and length of leaves, is essential for improving yield and precision breeding. Despite advances in computer vision and deep learning, accurate phenotypic identification remains challenging in field conditions due to complex background and environmental factors. Current methods, which are mostly designed for controlled environments, struggle with these challenges.

A Study (DOI: 10.34133/plantfenomics.0199) published in plant phenomics 29 May 2024 A point-line net model is proposed to improve field phenotypic identification by accurately detecting and tracking the position and trajectory of maize leaves.

In this study, the research assessed the object detection accuracy for maize using three popular models: Faster R-CNN, RetinaNet, and YOLOv3. Using the original model architecture, it was found that Faster R-CNN with ResNet101 + FPN achieved the highest performance with a MAP50 of 76.2% and a MAP75 of 39.9%, albeit with a longer detection time of 89.6 ms. with. To increase the accuracy, hyperparameters were fine-tuned, and soft-NMS and D IOU techniques were incorporated, improving MAP50 to 75.5% and MAP75 to 49.2%. Inspired by human keypoint detection, the research developed an innovative point-line net model, which achieved MAP50 of 81.5% and MAP75 of 50.1%, outperforming traditional methods. This method demonstrated superior accuracy in describing leaf and petiole trajectories with a custom distance assessment index (MLD) of 33.5, indicating its effectiveness in complex field environments. The training and validation process revealed that the model stabilized around the 100th epoch, suggesting optimal performance for subsequent prediction tasks.

According to Xu Ruan, the study’s lead researcher, “We believe the results of this study may also provide ideas for field management and phenotypic data collection for other crops.”

In summary, the point-line net model achieved an object detection accuracy (MAP50) of 81.5% and introduced a new lightweight keypoint detection branch, which significantly improved phenotypic detection. This research highlights the potential of deep learning methods to increase the efficiency of field plant phenotyping, providing valuable insights for future crop breeding and management. Integrating additional annotation information such as specific growth stages and multi-angle data can further increase the accuracy and applicability of models, paving the way for more accurate farming practices and better crop yield predictions.

,

Reference

DOI

10.34133/plantphenomics.0199

original source url

https://doi.org/10.34133/plantfenomics.0199

Author

bingwen liu^1,2,†Jianye Chang^2,†Dengfeng Hou^1,2Yuchen Pan^1,2dengao li^1,*Ju Ruan^2,**

Affiliation

1College of Computer Science and Technology (College of Data Science), Taiyuan University of Technology, Taiyuan, 030024, Shanxi, China.

2 Shenzhen Branch, Lingnan Guangdong Laboratory for Modern Agriculture, Genome Analysis Laboratory of the Ministry of Agriculture and Rural Affairs, Institute of Agricultural Genomics in Shenzhen, Chinese Academy of Agricultural Sciences, 518120, Shenzhen, China.

* Address correspondence: (email protected).

** Correspondence address: (email protected).

† These authors contributed equally to this work.

Funding Information

This work was supported by the Genome Refinement of the Main Model Organism and its Demonstration and Application-Subtopic 1 (2022YFC3400300), Acquisition and Decoding of Current Signals for Biological Nanopore Sequencing-Subtopic (2019YFA0707003), and the Agricultural Science and Technology Innovation Program .

About this plant phenomics

plant phenomics is an open access journal published in collaboration with the State Key Laboratory of Crop Genetics and Germplasm Cultivation, Nanjing Agricultural University (NAU) and published by the American Association for the Advancement of Science (AAAS). Like all partners participating in the Science Partner Journal program, plant phenomics Editorially it is independent from the Science family of magazines. Editorial decisions and scientific activities adopted by the Journal’s Editorial Board are carried out independently, on the basis of scientific merit and adhering to the highest standards for the accurate and ethical promotion of science. These decisions and activities are in no way influenced by the financial support of NAU, the NAU administration, or any other institutions and sponsors. The Editorial Board is solely responsible for all content published in the magazine. To learn more about the Science Partner Journal Program, visit the SPJ program homepage.