How to Grade Flowers Using Orchids as an Example?
Importance of Flower Grading
Phalaenopsis orchids, known for their captivating beauty, dominate the global orchid market, accounting for 79% of sales. In Asia alone, 84 million potted Phalaenopsis are sold annually, with China leading the charge.
But beneath this floral elegance lies a tedious task — manual flower counting for quality grading.
Growth and development of Phalaenopsis at different nitrogen and phosphorus levels. Source: Ruamrungsri et al., 2007
In the figure above: T1: 100:50, T2: 150:50, T3: 200:50, T4: 100:100, T5: 150:100, T6: 200:100 N:P.
Images of 22 Phalaenopsis species. Source: Tsai et al, 2015https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0141761
This process is not only labour-intensive but fraught with inaccuracies. Addressing these challenges, researchers have introduced an innovative approach: a combination of advanced deep learning and multiviewpoint imaging to transform flower grading.
Flowchart of the related work of our study. Source: Yang et al., 2024
In the figure above: (a) preparation of image data; (b) flower detection; and (c) flower counting.
##A Visionary Approach: PA-YOLO and Multiviewpoint Imaging
At the core of this breakthrough is the PA-YOLO algorithm, a tailored version of the YOLOv5 object detection framework. Enhanced for precision and robustness, PA-YOLO is designed to excel where traditional methods fall short—handling complex occlusions and overlapping blooms.
Example images of a potted Phalaenopsis sample. Source: Yang et al., 2024
In the figure above: (a) a collection of images from six viewpoints; (b) an example of an annotation, where the yellow, red and green rectangular boxes correspond to the classes of normal blooms, occluded blooms and buds, respectively; (c) an example of occlusion, where the red rectangular box with an arrow indicates a slightly occluded bloom and the red boxes without an arrow denote severely occluded blooms; and (d) an example of overlap, where the red circle highlights the only visible edge of a bloom that overlaps with a normal bloom marked by a yellow box.
Model structure of YOLOv5s - a, and PA-YOLO - b. Source: Yang et al., 2024
Key elements of the methodology include:
- Multiviewpoint Imaging System: A state-of-the-art setup featuring a rotating platform, industrial camera, and control unit that captures 360-degree images of potted Phalaenopsis. This setup ensures no bloom is left unnoticed, even those partially hidden.
- Data-Driven Innovation: The research employed a rich dataset of 11,880 images, with rigorous annotation to classify flowers as buds, normal blooms, or occluded blooms. This meticulous preparation laid the foundation for robust model training.
- Algorithmic Enhancements: PA-YOLO integrates cutting-edge features: A two-scale detection branch to focus on the medium and small targets common in floral arrangements. Optimized bottlenecks for improved feature representation. The DyHead framework, which uses attention mechanisms to enhance the detection of occluded flowers.
Conveyor system with a detection chamber to count Phalaenopsis plants. A - the overall structure of the circular conveyer; B - the internal configuration of the detection chamber.Source: Yang et al., 2024
High-Tech Meets High-Accuracy
The results are nothing short of remarkable. PA-YOLO achieved a mean average precision (mAP) of 95.4% and an average precision (AP) of 91.9% for occluded blooms, outperforming traditional algorithms by a significant margin. The research identified that using three viewpoints yielded the best results, with a bloom counting accuracy of 95.56% and a bud counting accuracy of 96.25%.
But numbers alone don’t tell the full story. The PA-YOLO system doesn’t just detect flowers —it interprets complex visual scenarios, distinguishing between overlapping blooms and partially hidden flowers with unprecedented clarity.
This innovation transforms what was once a painstaking manual task into a seamless, automated process.
Grad-CAM of the same detected target of four models. Source: Yang et al., 2024
In the figure above: (a) YOLOv5s; (b) YOLOv5s+2SDB; (c) YOLOv5s+2SDB+ONoB; and (d) YOLOv5s+2SDB+ONoB+DyHead.
The thermodynamic features of different colours indicate the regional attractiveness to the network, with the red areas indicating the greatest influence on the network. The influences gradually lessen as the colours transition from red to yellow and lastly to blue.
Practical Impact: A Blooming Future
This breakthrough has far-reaching implications for the ornamental plant industry:
- Efficiency Boost: By automating flower counting, growers can significantly reduce labour and time investments, reallocating resources to other value-adding activities.
- Accuracy Redefined: With a detection system robust enough to handle occlusions, errors in grading—once a costly issue — are minimized. Scalability and Sustainability: Designed to integrate seamlessly into existing workflows, this system supports large-scale operations, enabling growers to meet rising market demands sustainably.
- Adaptable Beyond Orchids: While tailored for Phalaenopsis, the system’s adaptability makes it a viable solution for other crops requiring precise quality assessments.
Looking Ahead: A Model for Precision Agriculture
This research exemplifies how cutting-edge technology is reshaping agriculture, merging deep learning with practical applications to solve real-world challenges.
By addressing the unique needs of flower grading, the PA-YOLO system sets a new standard in precision agriculture, promising not just efficiency but a future where beauty and technology coexist seamlessly.
Stay tuned as these innovations continue to blossom, offering growers the tools they need to thrive in an increasingly automated and data-driven industry. Whether you’re a grower, technologist, or enthusiast, the future of flower grading is something to watch closely—and PA-YOLO is leading the charge.
Petiole Pro: Free Web-Tools for Quality Assurance of Ornamental Plants
Growing Beautiful Plants Made Simple
Love growing ornamental plants? We created some handy web tools to help you care for them better! Petiole Pro makes it easy to check how your plants are doing - from counting blooms to making sure they’re growing evenly.
Smart Tools That Do the Hard Work
No more guessing about your plant’s health! Our Petiole Pro tools use clever technology to help you spot things like new blooms and check if leaves are healthy. Whether you’re an expert grower or just starting out with your first plants, these tools are super easy to use.
Free for Everyone Who Loves Plants
We believe everyone should have access to great plant care tools. That’s why Petiole Pro is completely free! Petiole Pro is free until you reach commercial scale and need intensive image processing resources. Whether you’re growing gorgeous orchids or tending to other beautiful plants, we’re here to help your farm thrive.
Petiole Pro Roses Counter is currently available for yellow roses but can be adjusted
Check the Petiole Pro Roses Counter at for images of yellow roses and for videos of yellow roses. Do you need to count different ornamental crops? Let us know at support@petioleapp.com
