An improved small object detection method based on Yolo V3

Cheng, Xianbao; Qiu, Guihua; Yu, Jiguo; Zhu, Zhengming

doi:10.1007/s10044-021-00989-7

Cited by 29 publications

(11 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Small target detection can be enhanced by Xianbao, C., [25] proposing an improved technique based on YoloV3. Starting with the feature map acquisition network, we've made a few tweaks to it.…”

Section: ░ 2 Related Workmentioning

confidence: 99%

YOLO Based Deep Learning Model for Segmenting the Color Images

Rasi¹,

AntoBennet

Renjith

et al. 2023

IJEER

View full text Add to dashboard Cite

The first stage is to extract fine details from a picture using Red Green Blue (RGB) colour space is colour image segmentation. Most grayscale and colour picture segmentation algorithms use original or updated fuzzy c-means (FCM) clustering. However, due to two factors, the majority of these methods are inefficient and fail to produce the acceptable segmentation results for colour photos. The inclusion of local spatial information often results in a high level of computational complexity due to the repetitive distance computation between clustering centres and pixels within a tiny adjacent window. The second reason is that a typical neighbouring window tends to mess up the local spatial structure of images. Color picture segmentation has been improved by introducing Deep Convolution Neural Networks (CNNs) for object detection, classification and semantic segmentation. This study seeks to build a light-weight for object detector that uses a depth and colour image from a publically available dataset to identify objects in a scene. It's likely to output in the depth way by expanding the YOLO network's network architecture. Using Taylor based Cat Salp Swarm algorithm (TCSSA), the weight of the suggested model is modified to improve the accuracy of region extraction findings. It is possible to test the detector's efficacy by comparing it to various datasets. Testing showed that the suggested model is capable of segmenting input into multiple metrics using bounding boxes. The results shows that the proposed model achieved 0.20 of Global Consistency Error (GCE) and 1.85 of Variation of Information (VOI) on BSDS500 dataset, where existing techniques achieved nearly 1.96 to 1.86 of VOI and 0.25 to 0.22 of GCE for the same dataset.

show abstract

Section: ░ 2 Related Workmentioning

confidence: 99%

YOLO Based Deep Learning Model for Segmenting the Color Images

Rasi¹,

AntoBennet

Renjith

et al. 2023

IJEER

View full text Add to dashboard Cite

show abstract

“…The exact model creates no false positives and only detects important things (FP) [18]. i i (2) The F-1 score, as stated in Eq., is the harmonic mean of the accuracy and recall scores (3).…”

Section: A Model Evaluationmentioning

confidence: 99%

“…Using the input picture as training data, the convolution network may successfully learn the key characteristics of the recognised item. Repeated training steadily boosts the training model's performance to provide accurate target detection outcomes [2].…”

Section: Introductionmentioning

confidence: 99%

Investigation of You Only Look Once Networks for Vision-based Small Object Detection

Yang¹

2023

IJACSA

View full text Add to dashboard Cite

Small object detection is a challenging issue in computer vision-based algorithms. Although various methods have been investigated for common objects including person, car and others, small object are not addressed in this issue. Therefore, it is necessary to conduct more researches on them. This paper is focused on small object detection especially jewellery as current object detection methods suffer from low accuracy in this domain. This paper introduces a new dataset whose images were taken by a web camera from a jewellery store and data augmentation procedure. It comprises three classes, namely, ring, earrings, and pendant. In view of the small target of jewellery and the real-time detection, this study adopted the You Only Look Once (Yolo) algorithms. Different Yolo based model including eight versions are implemented and train them using our dataset to address most effective one. Evaluation criteria, including accuracy, F1 score, recall, and mAP, are used to evaluate the performance of the various YOLOv5, YOLOv6, and YOLOv7 versions. According to the experimental findings, utilizing YOLOv6 is significantly superior to YOLOv7 and marginally superior to YOLOv5.

show abstract

“…Nevertheless, the proposed model overlooks the high cost of training while pursuing enhanced detection performance. Xianbao et al [30] propose an improved algorithm based on YOLOv3. Three improvements are proposed: image segmentation and upsampling, bilateral scaling, and increasing the residual network element to reduce the feature loss and avoid gradient fading.…”

Section: Related Workmentioning

confidence: 99%

SODCNN: A Convolutional Neural Network Model for Small Object Detection in Drone-Captured Images

Meng,

Zhou,

Liu

2023

Drones

View full text Add to dashboard Cite

Drone images contain a large number of small, dense targets. And they are vital for agriculture, security, monitoring, and more. However, detecting small objects remains an unsolved challenge, as they occupy a small proportion of the image and have less distinct features. Conventional object detection algorithms fail to produce satisfactory results for small objects. To address this issue, an improved algorithm for small object detection is proposed by modifying the YOLOv7 network structure. Firstly, redundant detection head for large objects is removed, and the feature extraction for small object detection advances. Secondly, the number of anchor boxes is increased to improve the recall rate for small objects. And, considering the limitations of the CIoU loss function in optimization, the EIoU loss function is employed as the bounding box loss function, to achieve more stable and effective regression. Lastly, an attention-based feature fusion module is introduced to replace the Concat module in FPN. This module considers both global and local information, effectively addressing the challenges in multiscale and small object fusion. Experimental results on the VisDrone2019 dataset demonstrate that the proposed algorithm achieves an mAP50 of 54.03% and an mAP50:90 of 32.06%, outperforming the latest similar research papers and significantly enhancing the model’s capability for small object detection in dense scenes.

show abstract

An improved small object detection method based on Yolo V3

Cited by 29 publications

References 13 publications

YOLO Based Deep Learning Model for Segmenting the Color Images

YOLO Based Deep Learning Model for Segmenting the Color Images

Investigation of You Only Look Once Networks for Vision-based Small Object Detection

SODCNN: A Convolutional Neural Network Model for Small Object Detection in Drone-Captured Images

Contact Info

Product

Resources

About