Automatic Vision Based Classification System Using DNN and SVM Classifiers
Vickneswari Durairajah, Suresh Gobee, Amgad Muneer · 2018
In this paper, we construct an automatic classification vision system that is designed to recognize Malaysian herbs that are typically used for medical or culinary purposes. The proposed system employs two classifiers, Support Vector machine (SVM) and Deep Neural Network (DNN). The two classifiers have been implemented using OpenCV-Python. For the training test SVM achieved 86.63% recognition accuracy and DNN (TensorFlow) achieved 98% recognition accuracy. For the real life testing SVM achieved 74.63% recognition accuracy and DNN achieved 93% recognition accuracy. In the proposed system a total of 1000 leaves were used. A total of 50 samples of herbs were collected for each class and they were divided into two datasets. The first dataset which consisted 60% of the herbs samples were used for the training purpose and the other dataset with 40% of the herbs samples were used for the testing purpose. The time taken for each recognition process was 4 seconds for SVM and 5 seconds for DNN classifier. Also, the proposed system is capable of identifying the herbs leaves even though they are wet, dried and deformed with a recognition accuracy of 52.50%. Finally, based on the experiments that were done, the system proved to be very efficient and accurate with the highest recognition rate being 98%. The results indicate that the techniques used in the proposed system are significantly efficient when compared to the various techniques employed in the existing literature.