Dynamic Object Recognition Using Evolving Deep Learning Detectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for orchard scouting, such as human inspection and automated imaging, are labor-intensive, inaccurate, and fail to account for the dynamic nature of growing crops, leading to inconsistent and non-uniform counting and decision-making during the growing season.
Innovation Solution
A method and system using a digital image series and an object detector series, including a seed object detector and deep learning object detectors, to recognize and track the evolution of target objects over time, enabling accurate and continuous orchard scouting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human inspection and tabulation are used to monitor crop growth, then labor costs are high and counting accuracy is poor, but the system is simple to implement
Solution Approach 1:
The patent replaces manual human inspection with an automated machine vision system using cameras and deep learning algorithms. The system captures images of crops and uses neural networks to automatically identify, count, and track crop development stages, eliminating the need for manual counting while significantly improving accuracy and consistency.
Solution Approach 2:
The patent introduces an image buffer as an intermediary component that stores captured images for processing. This buffer allows the system to manage image data efficiently, enabling accurate crop counting without requiring complex real-time processing of all images simultaneously, thus balancing accuracy with system complexity.
2Measurement precision
If automated imaging systems capture multiple sequential images and continuously process them, then crop detection accuracy improves, but data storage requirements and processing complexity increase
Solution Approach 1:
The patent implements a strategy where processed images are discarded after extraction, and only essential crop data is retained in a database. The image buffer manages sequential images by storing them temporarily for processing, then discarding them after use, preventing unbounded data accumulation while maintaining detection accuracy through multi-image analysis.
Solution Approach 2:
The patent extracts only the essential crop information (identification, count, development stage) from the sequential images and stores this extracted data in a database, while the raw image data is discarded after processing. This separation of data extraction and storage reduces the quantity of stored data while preserving detection accuracy.
3Adaptability or versatility
If discrete segmentation approaches with continuous human interaction are used to calibrate results, then system adaptability improves, but error accumulation and inconsistency increase
Solution Approach 1:
The patent implements a self-calibrating deep learning system that automatically adapts to different crop types and conditions without requiring continuous human interaction. The neural network learns from training data and automatically adjusts its parameters to improve detection accuracy, eliminating the need for manual calibration while maintaining high adaptability and consistency across different orchard conditions.
Solution Approach 2:
The patent incorporates feedback mechanisms where the deep learning model continuously learns from processed images and adjusts its detection algorithms accordingly. The system uses feedback from training data and performance metrics to refine its crop identification and counting accuracy, maintaining consistency without human intervention.
4Area of stationary object
If aerial systems with dense canopy coverage are used for crop monitoring, then broad area coverage is achieved, but fruit detection accuracy decreases due to leaf blocking
Solution Approach 1:
The patent transitions from aerial top-down imaging to ground-level multi-angle imaging. By positioning cameras at ground level and capturing images from multiple angles (including upward views through the canopy), the system achieves both broad coverage and accurate fruit detection. This dimensional change allows cameras to see fruits that are blocked from aerial perspectives.
Solution Approach 2:
The patent uses deep learning-based semantic segmentation to separate fruits from leaves and branches in the images. The neural network identifies and segments fruit regions even when partially obscured by foliage, enabling accurate detection and counting while maintaining broad orchard coverage through systematic image capture.
Data Source
AI summary
A method for performing machine vision recognition activities for a dynamic object is disclosed, which includes the steps of (i) creating a digital image series that includes a seed digital image set including digital images and at least one subsequent digital image set including digital images; (ii) creating an object detector series, said creating step (ii) comprising a) creating a seed object detector from the seed digital image set, the seed object detector comprising an architecture and a set of weights directed to recognition of target objects; and b) creating at least one deep learning object detector from the at least one subsequent digital image set and derived from said seed object detector, the deep learning object detector including a deep learning architecture and a set of weights trained for recognition of the evolution of the target objects over time.