Deconvolutional Single Shot Detector Research Articles

Surface defect detection of strip steel based on GT-CutMix augmentation algorithm and improved DSSD model

Abstract Nowadays, defect detection technology based on deep learning continuously increases the surface quality requirements of hot-rolled strip steel. However, due to limitations in industrial production, defect datasets often suffer from insufficient training samples and imbalanced categories. This paper proposes effective solutions, namely the GT-CutMix offline data augmentation algorithm and lightweight small sample defect detection models. The GT-CutMix augmentation algorithm significantly improves defect utilization by accurately sampling defect locations and integrating them into the original data set. We design the S-DSSD defect detection model by constructing a lightweight SI-MobileNet to replace the ResNet101 backbone of the Deconvolutional Single Shot Detector (DSSD) network. This can reduce the resource parameters and consumption. At the same time, it can speed up training and inference. To further improve the detection accuracy, we integrate the Pyramid Split Attention (PSA) attention mechanism into the prediction module of DSSD and construct the SA-DSSD model. Under the GT-CutMix augmentation algorithm, the mAP of S-DSSD and SA-DSSD models on X-SDD dataset are 76.83% and 78.63%, respectively. Meanwhile, the corresponding detection speeds are 45 FPS and 40 FPS, respectively. In addition, on the NEU-DET cross-dataset experiment, the mAP of the SA-DSSD model reaches 74.88%. Our methods are highly effective and generalized for small sample defect detection, which can provide selective solutions for specific needs such as high speed and precision in different industrial production scenarios.

MT-DSSD: multi-task deconvolutional single shot detector for object detection, segmentation, and grasping detection

A robot that picks and places the wide variety of items in a logistics warehouse must detect and recognize items from images and then decide which points to grasp. Our Multi-task Deconvolutional Single Shot Detector (MT-DSSD) simultaneously performs the three tasks necessary for this manipulation: object detection, semantic segmentation, and grasping detection. MT-DSSD is a multi-task learning (MTL) method based on DSSD that reduces the amount of computation and achieves high speed compared to when separate models perform each task. Evaluations using the Amazon Robotics Challenge dataset showed that our model has a better object detection and segmentation performance than comparable methods, and an ablation study showed that MTL could improve the accuracy of each task. Further, robotic experiments for grasping demonstrated that our model could detect the appropriate grasping point.

Insulator Fault Detection in Aerial Images based on the Mixed-grouped Fire Single-shot Multibox Detector

Abstract In a complex background, insulator fault is the main factor behind transmission accidents. With the wide application of unmanned aerial vehicle (UAV) photography, digital image recognition technology has been further developed to detect the position and fault of insulators. There are two mainstream methods based on deep learning: the first is the “two-stage” example for a region convolutional neural network and the second is the “one-stage” example such as a single-shot multibox detector (SSD), both of which pose many difficulties and challenges. However, due to the complex background and various types of insulators, few researchers apply the “two-stage” method for the detection of insulator faults in aerial images. Moreover, the detection performance of “one-stage” methods is poor for small targets because of the smaller scope of vision and lower accuracy in target detection. In this article, the authors propose an accurate and real-time method for small object detection, an example for insulator location, and its fault inspection based on a mixed-grouped fire single-shot multibox detector (MGFSSD). Based on SSD and deconvolutional single-shot detector (DSSD) networks, the MGFSSD algorithm solves the problems of inaccurate recognition in small objects of the SSD and complex structure and long running time of the DSSD. To resolve the problems of some target repeated detection and small-target missing detection of the original SSD, the authors describe how to design an effective and lightweight feature fusion module to improve the performance of traditional SSDs so that the classifier network can take full advantage of the relationship between the pyramid layer features without changing the base network closest to the input data. The data processing results show that the method can effectively detect insulator faults. The average detection accuracy of insulator faults is 92.4% and the average recall rate is 91.2%.

A real-time recognition method of static gesture based on DSSD

Gesture recognition is of great significance for human-machine interaction and it has broad application prospects. In order to improve the detection accuracy and speed, a real-time recognition method of static gesture based on Deconvolutional Single Shot Detector (DSSD) is proposed in this paper. We have improved the original DSSD network and the deconvolution module, used the K-means clustering algorithm to select the aspect ratios of the prior boxes to improve the detection accuracy. The detection accuracy of small data set is improved by introducing transfer learning method, and the influences of three different base networks on the DSSD network model are discussed. In order to verify the effectiveness of the proposed method, we compared it with the gesture recognition methods based on SSD300, SSD321, YOLOV2 and DES in ASL dataset. The experimental results show that the proposed method has a recognition rate of 94.8%, which is 2.7%, 2.1% and 2.8%higher than SSD300, SSD321 and YOLOv2, respectively. The detection rate is close to the method of Single-Shot Object Detection with Enriched Semantics (DES), while still maintaining a reasonable detection speed of 27 FPS. In addition, since DSSD fuse the semantic information of each feature extraction layer, the proposed method also has good detection ability for small gesture objects.

FFESSD: An Accurate and Efficient Single-Shot Detector for Target Detection

The Single Shot MultiBox Detector (SSD) is one of the fastest algorithms in the current target detection field. It has achieved good results in target detection but there are problems such as poor extraction of features in shallow layers and loss of features in deep layers. In this paper, we propose an accurate and efficient target detection method, named Single Shot Object Detection with Feature Enhancement and Fusion (FFESSD), which is to enhance and exploit the shallow and deep features in the feature pyramid structure of the SSD algorithm. To achieve it we introduced the Feature Fusion Module and two Feature Enhancement Modules, and integrated them into the conventional structure of the SSD. Experimental results on the PASCAL VOC 2007 dataset demonstrated that FFESSD achieved 79.1% mean average precision (mAP) at the speed of 54.3 frame per second (FPS) with the input size 300 × 300, while FFESSD with a 512 × 512 sized input achieved 81.8% mAP at 30.2 FPS. The proposed network shows state-of-the-art mAP, which is better than the conventional SSD, Deconvolutional Single Shot Detector (DSSD), Feature-Fusion SSD (FSSD), and other advanced detectors. On extended experiment, the performance of FFESSD in fuzzy target detection was better than the conventional SSD.

Object Detection Algorithm Based on Multiheaded Attention

This study proposes a multiheaded object detection algorithm referred to as MANet. The main purpose of the study is to integrate feature layers of different scales based on the attention mechanism and to enhance contextual connections. To achieve this, we first replaced the feed-forward base network of the single-shot detector with the ResNet–101 (inspired by the Deconvolutional Single-Shot Detector) and then applied linear interpolation and the attention mechanism. The information of the feature layers at different scales was fused to improve the accuracy of target detection. The primary contributions of this study are the propositions of (a) a fusion attention mechanism, and (b) a multiheaded attention fusion method. Our final MANet detector model effectively unifies the feature information among the feature layers at different scales, thus enabling it to detect objects with different sizes and with higher precision. We used the 512 × 512 input MANet (the backbone is ResNet–101) to obtain a mean accuracy of 82.7% based on the PASCAL visual object class 2007 test. These results demonstrated that our proposed method yielded better accuracy than those provided by the conventional Single-shot detector (SSD) and other advanced detectors.

FSNet: A Target Detection Algorithm Based on a Fusion Shared Network

Currently, in a target detection framework, a feature pyramid is widely used to capture differences resulting from the change in size of a detected object. Examples include the one-stage detector, the Deconvolutional Single-Shot Detector (DSSD), the RefineDet and the two-stage detector DetNet. Although these target detection frameworks with characteristic pyramid structure have achieved good results, they are limited by the facts that 1) they only use one feature layer of multi-scale information for prediction, and 2) they do not fully incorporate multi-scale inclusion of feature information. Therefore, based on Single-shot detector (SSD) architecture, we propose a novel one-stage target detector framework-fusion shared network (FSNet) which makes full use of feature information between multiple scale feature maps to more effectively detect objects of different scales. First, we introduce a fusion attention mechanism to fuse features between feature layers of different scales. Secondly, we combine this with multi-head structure fusion multi-layer features extracted as new features by a backbone network. Finally, we feed multi-layer fusion features into shared prediction module (shared PM). In the module, a new class feature pyramid for target detection is formed wherein the feature map used for prediction is composed of multiple levels of feature layers. In order to verify the effectiveness of this new FSNet target detection algorithm, we used the 320X320 input for FSNet, obtaining 80.7% mean accuracy (mAP) and 15 frames per second (Frame Per Second, FPS) in the PASCAL VOC 2007 test.