Area-Aware AI Model Loading for Object Recognition Control

The patent divides the object recognition system into multiple specialized AI models, each trained to recognize specific types of objects (e.g., food objects, pet objects). Instead of using one large comprehensive model that requires significant memory, the system segments the recognition task into multiple smaller specialized models that can be loaded selectively based on the recognition task at hand, thereby reducing the memory capacity required at any given time.

The system dynamically loads and switches between different AI models based on the current recognition needs. The processor selectively loads only the necessary model (e.g., food recognition model or pet recognition model) into volatile memory when needed, and can switch between models as different objects need to be recognized. This dynamic approach allows the system to maintain high recognition capability while managing limited memory resources efficiently.

The patent segments the complex object recognition task into multiple specialized AI models, where each model is optimized for recognizing specific object categories. This segmentation allows each model to be relatively simple and efficient, rather than requiring one massive model to handle all object types. The processor can then execute these smaller specialized models with lower processing power requirements while maintaining high accuracy for each specific recognition task.

The system loads and executes only the specific AI model needed for the current recognition task, rather than running all possible models simultaneously. For example, when a food object needs recognition, only the food recognition model is loaded and executed. This partial action approach significantly reduces the processing capability requirements at any given moment while maintaining high recognition accuracy for the relevant objects.

The system dynamically loads AI models into volatile memory based on the current operational context and recognition needs. When the electronic apparatus operates in a specific area (e.g., kitchen, living room), the processor selectively loads the corresponding AI model(s) for that area into volatile memory. When the area changes or recognition needs change, the system switches to loading different models. This dynamic loading approach maintains high adaptability across different areas while keeping volatile memory usage low at any given time.

The system prepares and stores multiple AI models in non-volatile memory in advance, organized by area or object type. Before actual recognition tasks are performed, the processor identifies which model is needed and loads it from non-volatile memory to volatile memory. This preliminary organization and selective loading approach allows the system to maintain adaptability for multiple areas without requiring all models to reside in volatile memory simultaneously, thereby managing memory resources efficiently.