Image classification is no longer a stranger in every sort of domain, especially in industries. This type of classification problem exists in various fields such as precision agriculture, control and sorting of automated crop tracking and more. The application of image classification has been used widely as it is capable of identifying products without any direct contact which can save human resources and time, which will increase the productivity of a business and lower its cost.
As there are plenty of techniques in image classification, three standard methods such as SIFT, SURF and CNN are being compared in this study to understand which of the methods is more effective in completing such tasks. Background study has been done to understand the applications of image classification in different domains and the techniques used, background regarding the standard methods such as SIFT, SURF and CNN.
To test out the techniques, we have introduced a fruit classification problem using the Fruit-360 dataset and construct the models using SIFT and SURF as feature extraction functions and SVM as the classifier. CNN is used to classify the fruits without any external feature extraction methods. Then, several evaluation methods are proposed to evaluate the efficiency of the models in classifying the classes of fruits. Based on the findings, CNN exhibits the most accurate classification of the fruits compared to the other two models, which are using SIFT and SURF as feature extraction methods.