Dietary assessment framework using improved K-means clustering and optimisation assisted CNN

Abstract

ABSTRACT Dietary intakes portray what someone consumes throughout a day, which also offer important insight for carrying out the intercession programs to avoid several chronic diseases. Precise tools and techniques of assessing nutritional intakes are necessary for monitoring the dietary condition of patients for clinical and epidemiological research on the relationships between health and diet. Measuring the exact dietary intakes is regarded as an open issue in the health and nutrition fields. Thereby, this work develops a new dietary assessment model implemented with four main phases like ‘(1) pre-processing, (2) segmentation, (3) feature extraction and (4) classification’. The input image is first subjected to pre-processing, which includes median filtering to decrease noise. The pre-processed image is then used in segmentation, which is performed using modified K-means clustering. Then comes feature extraction, which involves extracting colour, shape and texture-based features. Following that, the collected features are classified using an optimised CNN. Furthermore, the weights of CNN are fine-tuned using Self-Adaptive Cat Swarm Optimisation (SA-CSO), which is an upgraded version of the CSO algorithm, to improve the accuracy of the created approach.