A Remote Sensing Ship Recognition Using Random Forest

Abstract

In order to detect the marine targets reliably and timely, a novel ship recognition method by using optical remote sensing data based on random forest is presented. First, in the feature extraction part, in addition to the common features, we introduce the visual saliency features of the target.; second, an improved random forest based on mutual information (MIRF) is utilized to recognize ships in data from the optical remote sensing system; finally, we compare MIRF to classical algorithms. The MIRF has accelerated the operation speed of the algorithm and the classification accuracy remains robust. Theoretical analysis and experiment results show that the proposed method can achieve high recognition rate; therefore, this approach is feasible and efficient in the marine target recognition.

Imbalanced Remote Sensing Ship Image Classification

“…In order to verify the validity of the proposed method, we compare the WDNF with five representative ship classification methods: the k-nearest neighbor (KNN) [8] method, support vector machine (SVM) [9] method, affinity propagation (AP) [10], entropy-based hierarchical discriminant regression (E-HDR) [11] method, deep neural decision forest (dNDF), and MIRF [2]. dNDF [4] and WDNF adopt the same structure.…”

“…Because the final result is obtained by voting of each component classifier, the impact of unbalanced data can be reduced to a certain extent. Huang et al [2] extracted the conventional features and visual saliency features of ships, and then classified the high resolution remote sensing ship images using random forest based on mutual information (MIRF), which got good results. Unfortunately, random forests lack an effective mechanism for learning internal representations to help capture the main factors of data change [3].…”

Motion Pattern Classification on Online/Active Data-Machine Learning

“…The detection of ship patterns on offline data is possible by applying the mathematical algorithms. However, the problem statement is to detect online ship/vessel manoeuvring patterns in sea such as Zig-Zag, Loop, Parallel movement & sudden stop in mid sea for [2,3,4] Undergraduate Student, Department of Computer Science Engineering Sridevi Women's Engineering College, Hyderabad ujwalavure19@gmail.com gowthamirameshwaram@gmail.com Radar/AIS track data and The ship detection and pattern of the motion is observed at regular intervals based on time series for the above best classification technique. Finally, a smart object recognition system will be developed for the automatic motion pattern recognition of the vessel in videos/Frames.…”