Machine-Vision Produce Grading With Multi-View Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing agricultural produce grading and sorting systems face challenges due to natural variation in shape, size, and internal condition, leading to inconsistent manual inspection and the need for robust visual processing across multiple viewpoints and dimensions.
Innovation Solution
A system utilizing machine vision and robotics, including a robotic frame with actuated arms and a machine-vision system, captures multiple viewpoints of produce items, extracts object-level visual features, and controls the arms for precise handling operations based on defect classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection methods are used, then operational flexibility is maintained, but measurement precision and repeatability deteriorate
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated machine vision system comprising cameras, lighting, and image processing algorithms. The system captures images from multiple angles, processes them through computer vision algorithms to detect defects, and automatically classifies produce items, eliminating human subjectivity while maintaining operational simplicity through software-based decision rules.
Solution Approach 2:
The system creates visual copies (images) of the produce items from multiple viewpoints and uses these copies for analysis. By capturing and processing multiple images of each item, the system can detect subtle defects without physically manipulating the produce, thereby maintaining precision while keeping the physical handling mechanism simple.
2Measurement precision
If single-viewpoint imaging is used, then device complexity is reduced, but measurement precision and defect detection capability deteriorate
Solution Approach 1:
The patent transitions from single-viewpoint (2D) imaging to multi-viewpoint (3D) imaging by positioning cameras at different angles and heights. This dimensional expansion allows the system to detect defects on all surfaces of irregularly shaped produce items, significantly improving detection accuracy while the added complexity is managed through standardized camera mounting configurations.
Solution Approach 2:
The imaging system is segmented into multiple independent camera units, each capturing a specific viewpoint. This segmentation allows the system to comprehensively examine complex produce geometries by combining information from multiple sources, improving precision while enabling modular system design that manages overall complexity.
3Productivity
If automated handling is implemented, then productivity increases, but handling precision and produce integrity deteriorate
Solution Approach 1:
The patent replaces mechanical handling with vision-guided robotic manipulation. The machine vision system precisely locates each produce item's position, orientation, and features, then guides robotic arms to perform gentle, accurate handling operations. This substitution enables high-speed automated processing while maintaining produce integrity through precise, controlled movements that minimize mechanical stress.
Solution Approach 2:
The system implements closed-loop feedback by continuously monitoring produce position and orientation through machine vision, then adjusting robotic handling actions in real-time. This feedback mechanism ensures high throughput by enabling rapid automated decisions while protecting produce integrity by making real-time adjustments to prevent damage during handling and sorting operations.
Data Source
AI summary
A system including: a robotic frame including at least an actuated arm, a machine-vision system including at least a camera, the machine-vision system communicatively connected to a computing device, the computing device including at least a processor and a memory communicatively connected to the at least a processor, the memory storing instructions configuring the at least a processor to: receive image data from the at least a camera, the image data including at least two captured viewpoints of a produce item, extract at least an object-level visual feature from the image data using a machine vision model, assign a defect classification label to the produce item as a function of the at least an object-level visual feature, and control the at least an actuated arm to perform a handling operation on the produce item as a function of the defect classification label.