Food Classification by extracting the important features using VGGNet based Models in Precision Agriculture

Image categorization is quickly rising to the forefront as a potential component of computer vision-based object recognition. Nevertheless, study has only made preliminary progress in exploring the matter. So far, the superficial classification of food images has had a major impact on evaluating the nutritional capacities of individuals from different cultures based on the classification of their traditional cuisine. Preexisting algorithms already provide the most effective method for categorising all the different sorts of food. These devices have a limited capacity to recognise a specific quantity of meals within a given time frame. However, a single model has a finite capacity to identify objects. The chief objective of this work is to develop a recognition model that can classify different food items using transfer learning techniques. This study classified fruits and illnesses using a hybrid technique that combines contour feature-based classification with pre-trained networks. To extract valuable features from a plant dataset, a pretrained deep learning model that has already been fine-tuned (VGG19) was retrained. The next step was to use a hybrid method called HCSA-SFO, The system employs cuckoo search (CSA) and sunflower optimisation (SFO) as its basis to choose features with maximum efficiency. Subsequently, the researchers employed the pyramid histogram of oriented gradient (PHOG) technique to extract the contour characteristics of the object. Subsequently, the "Food-101" dataset, which is accessible to the public, was utilised to evaluate the Inception V3 model's classification capabilities. A model was created and tested using experimental data; it achieved a 97% accuracy degree in classifying 101 different types of food. From the analysis, it shows that the proposed model outperformed existing state-of-the-art representations in terms of accuracy.