Ship Detection from Highly Cluttered Images Using Convolutional Neural Network

Classification and Detection of vessels from cluttered images is an actively researched field of computer vision. The complexity of the coastal area makes it more complex in distinguishing different vessels. Environment conditions and visual aspect of the vessels make classification problem more difficult. The visual aspect ratio is very small and similar to most type of vessels. Since the traditional CNNs methods are slow and not much accurate, we present the robust method for the classification of ships derived from Support Vector Machine, bag of features and Convolutional Neural Networks (CNNs). First of all Support Vector Machine which is a supervised transfer leaning technique is introduced to remove the false candidates. As the ships have wide range of variability, bag of features are applied to handle the diversified features of different categories of vessels. Finally CNNs framework is applied for the deep feature extraction to successfully classify the ships from database. The proposed algorithm is trained with 16 different categories of ships with a data set of more than 2700 images. The algorithm trains and validates the results in terms of confusion matrix and provides better accuracy to classify the ships. The proposed Convolutional Neural Networks (CNNs) model outperforms the other previous methods and gives an accuracy of 91.04 percentages.

Automated summary

The study introduces a robust method for ship classification using a combination of Support Vector Machine, bag of features, and Convolutional Neural Networks, achieving a high accuracy of 91.04%. By integrating different algorithms and techniques, the system is able to effectively distinguish ships from cluttered images.