Animal Identification from Back Images Using Deep Metric Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing animal identification technologies, such as those relying on nose print images, face high-resolution image acquisition challenges, especially for moving animals like cows, necessitating a method that can identify individuals using alternative features.
Innovation Solution
A method utilizing deep metric learning models to determine embedding vectors from images of an animal's back, calculating distances between registered and target vectors, and identifying individuals based on these distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If nose print image matching is used for animal identification, then identification accuracy can be achieved, but high-resolution face images are required which are difficult to acquire from moving animals
Solution Approach 1:
The patent extracts the identification task from the face region and applies it to the back region of animals. By using back images instead of face images, the system avoids the difficulty of capturing high-resolution face images from moving animals while maintaining identification capability through feature extraction and embedding vector comparison
Solution Approach 2:
The patent creates a digital copy of the animal's back image and processes it through deep metric learning models to generate embedding vectors. This digital representation serves as a substitute for physical high-resolution imaging, allowing identification without requiring actual high-quality captured images of the animal's face or distinctive features
2Manufacturing precision
If high-resolution images are captured from moving animals, then detailed feature extraction is possible, but the animals must be stopped which reduces productivity
Solution Approach 1:
The patent replaces the mechanical approach of stopping animals for imaging with a computational approach. Instead of physically restraining or stopping animals to capture images, the system uses deep metric learning models to extract features from lower-resolution back images, eliminating the need for animal restraint and maintaining productivity
Solution Approach 2:
The patent changes the parameter of image resolution requirement from high-resolution to lower-resolution images. By modifying the input image quality parameter and using advanced computational models to compensate, the system achieves sufficient identification accuracy without requiring high-resolution captures that necessitate stopping moving animals
Data Source
AI summary
A method of the present disclosure includes (a) acquiring a target image related to a back of the target animal, (b) determining an embedding vector from the target image by using a deep metric learning model, (c) calculating a distance between the registered embedding vector and the embedding vector by using registered data including a registered embedding vector generated in advance for each of a plurality of registered individuals, and (d) determining an individual of the target animal from among the plurality of registered individuals by using the distance.