Sonar Image Detection Based on Multi-Scale Multi-Column Convolution Neural Networks

Abstract

Automatic detection of underwater objects by sonar images is an important and challenging topic in applications of Autonomous Underwater Vehicle (AUV) under the complex marine environment. A detection method is proposed based on Multi-Scale Multi-Column Convolution Neural Networks (MSMC-CNNs). Firstly, the Multi-Scale Multi-Column CNNs is used to form an encoder network for extracting multi-scale features of the sonar image. Secondly, the bicubic linear interpolation algorithm is used as the deconvolution process of the decoder networks to restore the sonar image size and resolution. Moreover, a novel transfer learning manner based on progressive fine-tuning to accelerate the model training. Finally, the proposed method is validated on the sonar image dataset and is compared with other existing detection methods. The pixel accuracy (PA) of MSMC-CNNs for different categories sonar image is over 95%. The experiment results show that the MSMC-CNNs model has better detection effect and more robustness to noise. INDEX TERMS Underwater object detection, MSMC-CNNs, bicubic linear interpolation algorithm, deconvolution. He received the B.S. degree in mathematics from Northwest University, China, in 1988, the M.S. degree in applied mathematics from Northwest Polytechnic University, China, in 1995, and the Ph.D. degree in electromagnetic field and microwave from Air Force Engineering University, China, in 2001. He is currently a Professor with Xijing University and a Visiting Scholar with the Department of Computer Science, Virginia Tech. His research interests include machine learning and its application in data mining, including machine learning, leaf image processing, data reduction, data mining, feature selection, wavelet transforms, and their application in the sonar image recognition.

Aircraft Detection in Remote Sensing Image based on Multi-scale Convolution Neural Network with Attention Mechanism

“…The traditional deep learning based image segmentation and detection models such as FCN [8], SegNet, U-Net and PSP-Net apply bilinear interpolation to recover image resolution during deconvolution [26]. Multi-scale multi-column CNN (MSMC-CNN) adopts the bicubic convolution as a deconvolution for the decoder structure to restore the original image size [27]. The results validate that MSMC-CNN is superior to the other segmentation methods.…”

“…The number of the aircraft RSIs of EORSSD is relatively small for training MSCNNA. To increase the image diversity, decrease the overfitting problem, and improve the aircraft detection ability, each original aircraft RSI is augmented by a rotation method (90 0 , 180 0 , 270 0 ) and a randomly cropping method with an overlap of 100 pixels, and then data cleaning is adopted to reduce sub-images without objects [27]. Each image is augmented to 4 images including aircraft targets with 3 orientations and a cropping, and then the augmented dataset contains 1548 images in total for aircraft detection task.…”

Sonar Image Target Detection and Recognition Based on Convolution Neural Network

“…e basic network used by YOLOv3 is Darknet53 [11]. To avoid the gradient explosion caused by the deepening of network layers, Darknet53 adds RESNET (residual neural network) [21][22][23][24][25] residual structure. When classifying the target objects, YOLOv3 uses several independent logistic regression classifiers.…”

Multi-Dimensional Underwater Point Cloud Detection Based on Deep Learning

“…The result represented that if the data was limited, the performance of transfer learning was better than that of the training convolutional neural networks (CNNs) from scratch. In [ 10 ], a CNN has been proven effective in extracting features from an image. A multi-scale and multi-column CNN was used to detect sonar images.…”