Comprehensive Performance Evaluation of YOLOv12, YOLO11, YOLOv10, YOLOv9 and YOLOv8 on Detecting and Counting Fruitlet in Complex Orchard Environments
This study provides a comprehensive comparative evaluation of YOLOv8 through YOLOv12 models for detecting and counting immature green fruitlets in complex orchard environments, finding that while YOLOv9 and YOLOv12 offer high detection accuracy, YOLO11n provides the best balance of counting precision and real-time inference speed.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The "Apple Eye" Challenge: Teaching Robots to Count Tiny Apples
Imagine you are standing in a massive, messy apple orchard. You are looking for tiny, green, baby apples (called "fruitlets") that are still growing. Here is the problem: these baby apples are the exact same color as the leaves, they are often hiding behind branches, and they are all huddled together in crowded clumps.
If a farmer wants to use a robot to help thin out these apples (removing some so the others can grow bigger and better), that robot needs a "brain" that can see through the chaos.
This scientific paper is essentially a "Battle of the Brains." Researchers took five of the world’s most advanced "AI eyes" (known as the YOLO family—which stands for "You Only Look Once") and put them through a grueling obstacle course to see which one is the best at spotting and counting these tiny green apples.
The Contestants: The YOLO Family
Think of the YOLO models like different types of athletes entering an obstacle course:
- YOLOv8: The "Old Reliable" veteran. Experienced and steady, but perhaps a bit slower than the new kids.
- YOLOv9: The "Smart Strategist." It uses clever math to make sure it doesn't lose important details while it's thinking.
- YOLOv10: The "Efficiency Expert." It’s designed to work fast by cutting out unnecessary steps.
- YOLO11: The "Speed Demon." It’s built for pure, lightning-fast reaction time.
- YOLOv12: The "High-Tech Specialist." It uses advanced "attention" mechanisms, like a human focusing intensely on one specific spot.
The Obstacle Course (The Test)
The researchers didn't just test these AI brains in a clean laboratory. They threw them into the "real world":
- The Camouflage Test: Can the AI tell the difference between a green apple and a green leaf?
- The Hide-and-Seek Test: Can it find an apple that is 90% covered by a branch?
- The Variety Test: Does it work on different types of apples (like Honeycrisp vs. Cosmic Crisp), or does it only recognize one kind?
- The Camera Swap: They trained the AI using a professional, expensive machine camera, but then tested it using a regular iPhone 14 Pro Max. This is like teaching a person to drive in a Ferrari and then asking them to drive a beat-up old truck—can they still do the job?
The Results: Who Won?
There wasn't just one winner; it depended on what you needed the robot to do:
- For Pure Accuracy (The Perfectionist): YOLOv9 was the champion of "Mean Average Precision." If you want the AI to be as accurate as possible in identifying exactly where the apples are, YOLOv9 is your best bet. It’s like a scholar who double-checks every single answer.
- For Speed (The Sprinter): YOLO11n won the race. It was incredibly fast, processing images in just 2.4 milliseconds. If you have a robot moving quickly through an orchard and you need it to make decisions instantly, YOLO11 is the winner. It’s the athlete that reacts before you even realize something happened.
- For Counting (The Accountant): When it came to actually counting the total number of apples without making mistakes, YOLO11n also took the gold medal. It was the most reliable at giving a "final tally" that matched what a human would count.
Why Does This Matter?
Right now, thinning apples is a back-breaking, manual job that relies on seasonal workers. As it becomes harder to find people to do this work, we need robots.
This paper proves that we don't need million-dollar specialized cameras to make this work; we can use high-quality smartphones and smart AI like YOLO11 to create robots that can "see" and "count" just like a human. This helps farmers grow better fruit, reduces labor shortages, and makes farming more high-tech and efficient.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.