| บทคัดย่อ |
การวิจัยนี้เพื่อพัฒนาระบบคัดแยกคุณภาพผลอะโวคาโดโดยใช้การเรียนรู้ของเครื่องและคอมพิวเตอร์วิทัศน์เพื่อตรวจจับตำหนิ 3 ประเภท ได้แก่ รอยแมลงเจาะ (Insect-Borer), รอยช้ำ (Bruise) และโรคจุด (Spot) ระบบประกอบด้วย กล้องถ่ายภาพความละเอียดสูงและบอร์ด Raspberry Pi 4 Model B ควบคุมการประมวลผล และคัดแยกผลผลิตอัตโนมัติผ่านสายพานลำเลียงโมเดลถูกพัฒนาโดยใช้ชุดข้อมูลภาพที่ผ่าน Data Labeling และ Image Augmentation พร้อมฝึกสอนด้วยอัลกอริทึม YOLOv5s และ YOLOv5n ที่ 100, 200 และ 300 Epochs ผลการทดลองพบว่า YOLOv5n มีประสิทธิภาพสูงสุด และถูกนำไปใช้งานจริง โดยติดตั้งระบบ Push Rejector เพื่อคัดแยกอะโวคาโดตามประเภทที่ตรวจพบถูกส่งไปยังเอาต์พุตตามกำหนดได้โดยอัตโนมัติ การทดสอบบนชุดข้อมูลคลาสละ 50 ตัวอย่าง พบว่าโมเดลทำงานได้ดีที่สุดในคลาส Insect-Borer ค่า F1-score 93.62% และ Bruise F1-score 91.30% ซึ่งมีความแม่นยำสูง Recall 100% ค่าความถูกต้องโดยรวม (Accuracy) อยู่ที่ 74.67% แสดงให้เห็นว่าโมเดลมีประสิทธิภาพปานกลางถึงสูง ชุดระบบผลักชิ้นงานสามารถทำงานได้อย่างถูกต้อง สามารถตรวจจับรอยตำหนิบนผิวเปลือกของผลอะโวคาโดและคัดแยกผลผลิตที่มีตำหนิออกจากผลผลิตที่มีคุณภาพได้อย่างมีประสิทธิภาพ
This research aims to develop a quality sorting system for avocado fruit using machine learning and computer vision to detect three types of defects: insect-borer, bruise, and spot. The model was developed using image datasets that have undergone Data Labeling and Image Augmentation and trained with YOLOv5s and YOLOv5n algorithms at 100, 200, and 300 epochs, the results of the experiment showed that YOLOv5n was the most effective and was put into practical use by installing a push rejector system to sort the avocados according to the type detected and automatically sent to the output as scheduled. Testing on a dataset of 50 samples per class, it was found that the model performed best in the Insect-Borer class with an F1-score of 93.62% and a Bruise F1-score of 91.30%, which had a high accuracy of 100% recall, and an overall accuracy of 74.67%. The unit of the workpiece pushing system can work properly. It can detect blemishes on the peel surface of avocados and effectively sort out defective yields from quality yields.
|
| คำสำคัญ |
การเรียนรู้ของเครื่อง,การคัดคุณภาพ,คอมพิวเตอร์วิทัศน์,อะโวคาโด,Machine Learning,Quality grading,Computer Vision,Avocado
|