Inner-Imaging Networks: Put Lenses Into Convolutional Structure

Abstract

Despite the tremendous success in computer vision, deep convolutional networks suffer from serious computation costs and redundancies. Although previous works address that by enhancing the diversities of filters, they have not considered the complementarity and the completeness of the internal convolutional structure. To respond to this problem, we propose a novel inner-imaging (InI) architecture, which allows relationships between channels to meet the above requirement. Specifically, we organize the channel signal points in groups using convolutional kernels to model both the intragroup and intergroup relationships simultaneously. A convolutional filter is a powerful tool for modeling spatial relations and organizing grouped signals, so the proposed methods map the channel signals onto a pseudoimage, like putting a lens into the internal convolution structure. Consequently, not only is the diversity of channels increased but also the complementarity and completeness can be explicitly enhanced. The proposed architecture is lightweight and easy to be implement. It provides an efficient self-organization strategy for convolutional networks to improve their performance. Extensive experiments are conducted on multiple benchmark datasets, including CIFAR, SVHN, and ImageNet. Experimental results verify the effectiveness of the InI mechanism with the most popular convolutional networks as the backbones.

Defects Detection on 110 MW AC Wind Farm’s Turbine Generator Blades Using Drone-Based Laser and RGB Images with Res-CNN3 Detector

“…DarkNet-53 is a deep learning algorithm with high detection accuracy that is gaining prevalence in utilization for state-of-the-art object detection challenges [20]. InI-WRN-16-8-square-3 is a wide residual network backbone application of the InI module that includes square G-filters for CNN structure abstraction [21,22]. The InI-PyramidNet-mix-5 CNN technique is the result of applying the InI module to the pyramidal residual network backbone [21,23,24], and it includes mixed G-filters for structural modeling of CNNs.…”

75MW AC PV Module Field Anomaly Detection Using Drone-Based IR Orthogonal Images With Res-CNN3 Detector

“…This is a deep network architecture that can recognize probable destinations and learn destination-specific trajectories that outperforms existing state-of-the-art models. The Inner-imaging (InI) mechanism proposed in [25] is a technique for modeling the channel relationships in CNNs. It builds inner-imaged maps with both residual and identity mappings, enabling identification flows to participate in the attentional process of residual flows.…”

“…DarkNet-53 is a deep learning algorithm that is beginning to gain traction in applications for highly advanced object detection problems and it has very effective detection accuracy [20]. InI-WRN-16-8-square-3 is an application of the InI module on the wide residual network backbone, which comprises square Gfilters for CNN structure modelling [25,26]. InI-PyramidNetmix-5 is a resultant CNN model from the application of the InI module on the pyramidal residual network backbone [25,27], it comprises mixed G-filters for structural modelling of CNNs.…”

“…InI-WRN-16-8-square-3 is an application of the InI module on the wide residual network backbone, which comprises square Gfilters for CNN structure modelling [25,26]. InI-PyramidNetmix-5 is a resultant CNN model from the application of the InI module on the pyramidal residual network backbone [25,27], it comprises mixed G-filters for structural modelling of CNNs. Additionally, GoogleNet [21] and YOLO [22] are experimented with and assessed for PV module anomaly detection.…”

Chinese Character Components Segmentation Method Based on Faster RCNN

“…In recent years, neural networks [14], [15] have developed rapidly, and the research on Chinese characters has achieved more results with the rise of neural networks. For example, Zeng et al [16] and Zhang et al [17] applied neural networks to generate Chinese characters, and Zeng et al [16] also used stroke encoding information to ensure the quality of the generated Chinese characters.…”