Abstract
The deployment of deep convolutional neural networks (CNNs) in synthetic aperture radar (SAR) ship real-time detection is largely hindered by huge computational cost. In this article, we propose a novel learning scheme for training a lightweight ship detector called Tiny YOLO-Lite, which simultaneously 1) reduces the model storage size; 2) decreases the floating point operations (FLOPs) calculation; and 3) guarantees the high accuracy with faster speed. This is achieved by self-designed backbone structure and network pruning, which enforces channel-level sparsity in the backbone network and yields a compact model. In addition, knowledge distillation is also applied to make up for the performance decline caused by network pruning. Hereinto, we propose to let small student model mimic cumbersome teacher's output to achieve improved generalization. Rather than applying vanilla full feature imitation, we redefine the distilled knowledge as the inter-relationship between different levels of feature maps and then transfer it from the large network to a smaller one. On account that the detectors should focus more on the salient regions containing ships while background interference is overwhelming, a novel attention mechanism is designed and then attached to the distilled feature for enhanced representation. Finally, extensive experiments are conducted on SSDD, HRSID, and two large-scene SAR images to verify the effectiveness of the thinner SAR ship object detector in comparison of with other CNN-based algorithms. The detection results demonstrate that the proposed detector can achieve lighter architecture with 2.8-M model size, more efficient inference (>200 fps) with low computation cost, and more accurate prediction with knowledge transfer strategy.
Highlights
N OWADAYS, deep convolutional neural networks (CNNs) have gained much attention in the application of synthetic aperture radar (SAR) field, such as automatic target recognition [1], urban interpretation [2], marine surveillance [3], and Manuscript received September 20, 2020; revised November 18, 2020; accepted November 26, 2020
In this article, we proposed to learn a simple but efficient SAR ship detector through the combination of network pruning and knowledge distillation (KD), which would facilitate the deployment of SAR ship detection in practical applications and thereby improve the model generalization ability
2) Different from the most network pruning approaches which are attached with fine-tuning stage to make up for performance decline, we present a novel KD strategy for model deployment on spaceborne or missile-borne platform, which eliminates the temporary performance degradation and further boost the detection performance
Summary
N OWADAYS, deep convolutional neural networks (CNNs) have gained much attention in the application of synthetic aperture radar (SAR) field, such as automatic target recognition [1], urban interpretation [2], marine surveillance [3], and Manuscript received September 20, 2020; revised November 18, 2020; accepted November 26, 2020. Feature concatenation, and anchor box mechanism are adopted to improve detection accuracy, their model still involves partial heavy convolution operators, declining the detection speed They further proposed HyperLi-Net [20] by integrating five internal mechanisms and five external modules, which is more lightweight and accurate. To further improve the detection performance of pruned models, we propose feature map inter-relationship guided KD, a novel model compression approach for efficiently training a slim but effective ship detector. The problem becomes tougher especially when detecting inshore ships in SAR images since the interference of backscattering points are stronger and many complex surroundings may cause more false alarms To address this dilemma, a novel graph-based feature KD strategy is devised to perform imitations at the most discriminative features from different levels. A novel attention-based block is proposed to strengthen distilled features according to the object-related regions, which largely eliminates the interference of complex cluttered background
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have