Computer-implemented method to increase efficiency in image classification (Machine-translation by Google Translate, not legally binding)

• Main Authors: VILLALBA ORERO, María, LE CLAINCHE MARTINEZ, Soledad, LARA PEZZI, Enrique, VALERO SANCHEZ, Eusebio, GROUN, Nourelhouda, GARICANO MENA, Jesús
• Format: Patent
• Language: English; Spanish
• Published: 09.07.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATIONTECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING ORPROCESSING OF MEDICAL OR HEALTHCARE DATA; INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTEDFOR SPECIFIC APPLICATION FIELDS; PHYSICS

Computer-implemented method for increasing efficiency in image classification that comprises: obtaining a set of input images (11); obtain at least a first input tensor associated with at least a subset of the set of input images (12), perform a decomposition in higher order dynamic modes (HODMD) of the at least one first input tensor (13), obtaining a first tensor of higher order dynamical modes (14); obtain at least a first expanded set associated with at least a subset of the input image set (15). After partitioning the expanded (or enriched, or augmented) set (16), a convolutional neural network (CNN) (7') is trained using at least a subset of the expanded set, thus obtaining a classification tool (8') capable of classifying new images (9') associated with the input images. (Machine-translation by Google Translate, not legally binding) Método implementado por ordenador de aumento de la eficiencia en la clasificación de imágenes que comprende: obtener un conjunto de imágenes de entrada (11); obtener al menos un primer tensor de entrada asociado a al menos un subconjunto del conjunto de imágenes de entrada (12), realizar una descomposición en modos dinámicos de orden superior (HODMD) del al menos un primer tensor de entrada (13), obteniendo un primer tensor de modos dinámicos de orden superior (14); obtener al menos un primer conjunto ampliado asociado a al menos un subconjunto del conjunto de imágenes de entrada (15). Tras la partición del conjunto expandido (o enriquecido, o aumentado) (16), se entrena una red neuronal convolucional (CNN) (7') utilizando al menos un subconjunto del conjunto expandido, obteniendo así una herramienta de clasificación (8') capaz de clasificar nuevas imágenes (9') asociado con las imágenes de entrada.