Tiny Targets Research Articles

Enhanced YOLOv8-Based Model with Context Enrichment Module for Crowd Counting in Complex Drone Imagery

Crowd counting in aerial images presents unique challenges due to varying altitudes, angles, and cluttered backgrounds. Additionally, the small size of targets, often occupying only a few pixels in high-resolution images, further complicates the problem. Current crowd counting models struggle in these complex scenarios, leading to inaccurate counts, which are crucial for crowd management. Moreover, these regression-based models only provide the total count without indicating the location or distribution of people within the environment, limiting their practical utility. While YOLOv8 has achieved significant success in detecting small targets within aerial imagery, it faces challenges when directly applied to crowd counting tasks in such contexts. To overcome these challenges, we propose an improved framework based on YOLOv8, incorporating a context enrichment module (CEM) to capture multiscale contextual information. This enhancement improves the model’s ability to detect and localize tiny targets in complex aerial images. We assess the effectiveness of the proposed framework on the challenging VisDrone-CC2021 dataset, and our experimental results demonstrate the effectiveness of this approach.

Small Target Detection Algorithm Based on Improved YOLOv5

Small targets exist in large numbers in various fields. They are broadly used in aerospace, video monitoring, and industrial detection. However, because of its tiny dimensions and modest resolution, the precision of small-target detection is low, and the erroneous detection rate is high. Therefore, based on YOLOv5, an improved small-target detection model is proposed. First, in order to improve the number of tiny targets detected while enhancing small-target detection performance, an additional detection head is added. Second, involution is used between the backbone and neck to increase the channel information of feature mapping. Third, the model introduces the BiFormer, wherein both the global and local feature information are captured simultaneously by means of its double-layer routing attention mechanism. Finally, a context augmentation module (CAM) is inserted into the neck in order to maximize the structure of feature fusion. In addition, in order to consider among the required real frame as well as the prediction frame simultaneously, YOLOv5’s original loss function is exchanged. The experimental results using the public dataset VisDrone2019 show that the proposed model has P increased by 13.43%, R increased by 11.28%, and mAP@.5 and mAP@[.5:.95] increased by 13.88% and 9.01%, respectively.

Moving hearts and minds: how the Big Tobacco, Tiny Targets campaign builds support for tobacco control policy

Moving hearts and minds: how the Big Tobacco, Tiny Targets campaign builds support for tobacco control policy

Impact of a national tsetse control programme to eliminate Gambian sleeping sickness in Uganda: a spatiotemporal modelling study.

Tsetse flies (Glossina) transmit Trypanosoma brucei gambiense, which causes gambiense human African trypanosomiasis (gHAT). As part of national efforts to eliminate gHAT as a public health problem, Uganda implemented a large-scale programme of deploying Tiny Targets, which comprise panels of insecticide-treated material which attract and kill tsetse. At its peak, the programme was the largest tsetse control operation in Africa. Here, we quantify the impact of Tiny Targets and environmental changes on the spatial and temporal patterns of tsetse abundance across North-Western Uganda. We leverage a 100-month longitudinal dataset detailing Glossina fuscipes fuscipes catches from monitoring traps between October 2010 and December 2019 within seven districts in North-Western Uganda. We fitted a boosted regression tree (BRT) model assessing environmental suitability, which was used alongside Tiny Target data to fit a spatiotemporal geostatistical model predicting tsetse abundance across our study area (~16 000 km2). We used the spatiotemporal model to quantify the impact of Tiny Targets and environmental changes on the distribution of tsetse, alongside metrics of uncertainty. Environmental suitability across the study area remained relatively constant over time, with suitability being driven largely by elevation and distance to rivers. By performing a counterfactual analysis using the fitted spatiotemporal geostatistical model, we show that deployment of Tiny Targets across an area of 4000 km2 reduced the overall abundance of tsetse to low levels (median daily catch=1.1 tsetse/trap, IQR=0.85-1.28). No spatial-temporal locations had high (>10 tsetse/trap/day) numbers of tsetse compared with 18% of locations for the counterfactual. In Uganda, Tiny Targets reduced the abundance of G. f. fuscipes and maintained tsetse populations at low levels. Our model represents the first spatiotemporal geostatistical model investigating the effects of a national tsetse control programme. The outputs provide important data for informing next steps for vector control and surveillance.

Patterns of microbiome composition in tsetse fly Glossina palpalis palpalis during vector control using Tiny Targets in Campo, South Cameroon.

Novel vector control tools against African trypanosomiases require a deep understanding of the factors driving tsetse vector fitness or population resilience in their ecosystems. Following evidence of microbiota-mediated host fitness or traits shaping, including insecticide resistance in arthropod populations, we undertook a comparative study of the microbiota in wild-caught tsetse flies during vector control with deltamethrin-impregnated traps called Tiny Targets. The bacterial microbiome composition of tsetse flies collected before and after 6, 12, and 18 months of vector control were characterized using high-throughput sequencing of the V3-V4 hypervariable region of the bacterial 16S rRNA gene and compared. Overall, 48 bacterial genera and five phyla were identified. The primary symbiont Wigglesworthia dominated almost all the samples with an overall relative abundance of 71.76%. A significant increase was observed in microbiome diversities over the vector control with new taxa identified. Interestingly, few genera, like Curvibacter for instance, displayed a regularly increasing abundance, from 0.57% to 0.65%, 4.73%, and 8.57% after 6, 12, and 18 months of tsetse control, respectively. This study provided preliminary for further investigation into the role and mechanism of action of microbiota in tsetse fly fitness under selective pressure like insecticides.IMPORTANCEThe interest in vector control in the fight against African trypanosomiases has been reinforced in recent years, with the development of small insecticide-impregnated screens, known as "Tiny Targets". As some tsetse biotopes are difficult to access for their installation, other tools are under consideration that involve using bacteria harbored by the tsetse vector to block the development of trypanosomes or impair the tsetse's fitness in its natural environment. Several bacterial symbionts were previously described as important for tsetse fly development, and some like Burkholderia and Citrobacter also found in tsetse flies were found associated with insecticide tolerance in other arthropods. In this research, we found the bacterial genera, Curvibacter and Acinetobacter, increased in abundance in tsetse flies during vector control. These bacteria deserve further attention to determine if they can interfere with insecticides used to control tsetse fly populations.

Subtle-YOLOv8: a detection algorithm for tiny and complex targets in UAV aerial imagery

Subtle-YOLOv8: a detection algorithm for tiny and complex targets in UAV aerial imagery

Role Exchange-Based Self-Training Semi-Supervision Framework for Complex Medical Image Segmentation.

Segmentation of complex medical images such as vascular network and pulmonary tracheal network requires segmentation of many tiny targets on each tomographic section of the 3-D medical image volume. Although semantic segmentation of medical images based on deep learning has made great progress, fully supervised models require a great amount of annotations, making such complex medical image segmentation a difficult problem. In this article, we propose a semi-supervised model for complex medical image segmentation, which innovatively proposes a bidirectional self-training paradigm, through dynamically exchanging the roles of teacher and student by estimating the reliability at the model level. The direction of information and knowledge transfer between the two networks can be controlled, and the probability distribution of the roles of teacher and student in the next stage will be jointly determined by the model's uncertainty and instability in the training process. We also resolve the problem that loosely coupled networks are prone to collapse when training on small-scale annotated data by proposing asymmetric supervision (AS) strategy and hierarchical dual student (HDS) structure. In particular, a bidirectional distillation loss combined with the role exchange (RE) strategy and a global-local-aware consistency loss are introduced to obtain stable mutual promotion and achieve matching of global and local features, respectively. We conduct detailed experiments on two public datasets and one private dataset and lead existing semi-supervised methods by a large margin, while achieving fully supervised performance at a labeling cost of 5%.

EMR-YOLO: A Study of Efficient Maritime Rescue Identification Algorithms

Accurate target identification of UAV (Unmanned Aerial Vehicle)-captured images is a prerequisite for maritime rescue and maritime surveillance. However, UAV-captured images pose several challenges, such as complex maritime backgrounds, tiny targets, and crowded scenes. To reduce the impact of these challenges on target recognition, we propose an efficient maritime rescue network (EMR-YOLO) for recognizing images captured by UAVs. In the proposed network, the DRC2f (Dilated Reparam-based Channel-to-Pixel) module is first designed by the Dilated Reparam Block to effectively increase the receptive field, reduce the number of parameters, and improve feature extraction capability. Then, the ADOWN downsampling module is used to mitigate fine-grained information loss, thereby improving the efficiency and performance of the model. Finally, CASPPF (Coordinate Attention-based Spatial Pyramid Pooling Fast) is designed by fusing CA (Coordinate Attention) and SPPF (Spatial Pyramid Pooling Fast), which effectively enhances the feature representation and spatial information integration ability, making the model more accurate and robust when dealing with complex scenes. Experimental results on the AFO dataset show that, compared with the YOLOv8s network, the EMR-YOLO network improves the mAP (mean average precision) and mAP50 by 4.7% and 9.2%, respectively, while reducing the number of parameters and computation by 22.5% and 18.7%, respectively. Overall, the use of UAVs to capture images and deep learning for maritime target recognition for maritime rescue and surveillance improves rescue efficiency and safety.

TFITrack: Transformer Feature Integration Network for Object Tracking

AbstractDue to the ignoring of rich spatio-temporal and global contextual information with convolutional neural networks in features extraction, the traditional method is prone to tracking drift or even failure in complex scenario, especially for the tiny targets in aerial photography scenario. In this work, it proposes a transformer feature integration network (TFITrack) to obtain diverse and comprehensive target feature for the robust object tracking. Based on the typical transformer architecture, it optimizes encoder and decoder structure for aggregating discriminative spatio-temporal information and global context-awareness feature. Furthermore, the encoder introduces the similarity calculation layer and dual-attention module; the aim is to deepen the similarity between features and make corrections for channel and spatial dimensions, and feature representation is improved. Finally, with the introduction of the temporal context filtering layer, unimportant feature information is ignored adaptively, obtaining a balance between the parameters number reduction and stable performance. Experimental results show that the proposed tracking algorithm exhibits excellent tracking performance on seven benchmark datasets, especially on the aerial dataset UAV123, UAV20L, and UAV123@10fps, which presents the advantages of the novel method in dealing with fast motion and external interference.

HV-YOLOv8 by HDPconv: Better lightweight detectors for small object detection

Accurately identifying and localising small objects within images or videos is a critical challenge in the field of computer vision. It is mostly applied in scenarios that require high real-time performance, such as pedestrian detection and autonomous driving scenarios. These tiny targets generally include small objects at long distances, or objects appearing in low-resolution images, due to which it becomes exceptionally difficult to extract effective feature information. Since YOLOv8 with its large downsampling multiplier leads to deeper feature maps that make it difficult to detect tiny objects, we find that the use of residual structures in the convolution module can enhance the accuracy of small object detection. However, this undoubtedly increases the computational cost, so we lightened the convolution module to make it more suitable for practical applications and named it Halved Deep Pointwise Convolution (HDPConv). A cross-level partial module Variety of View Group Shuffle Cross Stage Partial Network (VOV-GSCSP) is also utilised, using a rational architecture as well as multi-scale information fusion, to ensure that the overall model is lightweight while obtaining rich gradient flows. On this basis, we propose a new network lightweight model HV-YOLOv8. In multiple sets of comparative experiments on two datasets (containing several state-of-the-art solutions as well as classical ones), we demonstrate the superiority of HV-YOLOv8, in particular, the accuracy is improved by 1.4% compared to YOLOv8, while the number of parameters and the amount of computation are drastically reduced.

Research on Methods of Micro-target Detection in Various Application Scenarios

This paper provides a comprehensive introduction and comparison of the research on small target detection algorithms in different application scenarios and systematically explores the application of deep learning algorithms for small target detection in remote sensing imagery, visible imagery, and infrared imagery. In remote sensing images, the Deconvolution R-CNN algorithm is used to improve the accuracy of small target detection by adding a deconvolution layer. In the field of visible images, a Gaussian Mixture Model (GMM)-based approach is introduced to achieve fast detection of small targets with SIFT features and a GMM target detector. In infrared images, a comprehensive algorithm adapted to low signal-to-noise ratio and complex background is proposed to automatically detect tiny targets through background suppression and threshold selection optimization. In this paper, we compare and analyze the performance of these methods in different scenarios, and propose directions for future research, including expanding the training dataset, applying super-resolution methods to improve the image resolution, and enhancing the universality of the algorithms. This paper provides a comprehensive overview of research in the field of small target detection, as well as an outlook on future research directions, which provides a useful reference for the further development of the field.

Modelled impact of Tiny Targets on the distribution and abundance of riverine tsetse.

The insecticide-treated baits known as Tiny Targets are one of the cheapest means of controlling riverine species of tsetse flies, the vectors of the trypanosomes that cause sleeping sickness in humans. Models of the efficacy of these targets deployed near rivers are potentially useful in planning control campaigns and highlighting the principles involved. To evaluate the potential of models, we produced a simple non-seasonal model of the births, deaths, mobility and aging of tsetse, and we programmed it to simulate the impact of seven years of target use against the tsetse, Glossina fuscipes fuscipes, in the riverine habitats of NW Uganda. Particular attention was given to demonstrating that the model could explain three matters of interest: (i) good control can be achieved despite the degradation of targets, (ii) local elimination of tsetse is impossible if invasion sources are not tackled, and (iii) with invasion and target degradation it is difficult to detect any effect of control on the age structure of the tsetse population. Despite its simplifications, the model can assist planning and teaching, but allowance should be made for any complications due to seasonality and management challenges associated with greater scale.

National capacity strengthening within the context of an international vector control partnership: findings from a qualitative study conducted within the Ugandan ‘Tiny Targets’ programme

IntroductionThe Ugandan Tiny Target programme is an example of an international vector control partnership that held specific capacity strengthening objectives in support of a disease elimination goal. Drawing on this...

Real-Time Detection of Drill Pipe Joints Using Improved YOLOv5x Model Applied to Drilling Operation Images

It is widely known that pipe assembly and disassembly still lacks automation. To enhance drilling efficiency, we propose an autonomous detection and positioning model based on improved YOLOv5x for drill pipe joints. Firstly, we use Activate or Not (ACON) activation function and Convolution Block Attention Module (CBAM) to enhance feature extraction and representation ability. Then, the loss function is changed from Complete IoU (CIOU) loss to Scale-Invariant IoU (SIOU) loss, which increases the accuracy of the bounding box regression. Finally, the predict heads are tailored to be more effective in detecting tiny targets. Following network training, the final drill pipe joint detection model, based on the improved YOLOv5x, achieved an average accuracy of 99.10%, a 3.5% improvement over the base YOLOv5x algorithm. The proposed method effectively promotes the intelligence of oil drilling equipment by quickly and accurately detecting drill pipe joints, consequently enhancing drilling operation efficiency.

GWS: Rotation object detection in aerial remote sensing images based on Gauss–Wasserstein scattering

The majority of existing rotating target detectors inherit the horizontal detection paradigm and design the rotational regression loss based on the inductive paradigm. But the loss design limitation of the inductive paradigm makes these detectors hardly detect effectively tiny targets with high accuracy, particularly for large-aspect-ratio objects. Therefore, in view of the fact that horizontal detection is a special scenario of rotating target detection and based on the relationship between rotational and horizontal detection, we shift from an inductive to a deductive paradigm of design to develop a new regression loss function named Gauss–Wasserstein scattering (GWS). First, the rotating bounding box is transformed into a two-dimensional Gaussian distribution, and then the regression losses between Gaussian distributions are calculated as the Wasserstein scatter; By analyzing the gradient of centroid regression, centroid regression is shown to be able to adjust gradients dynamically based on object characteristics, and small targets requiring high accuracy detection rely on this mechanism, and more importantly, it is further demonstrated that GWS is scale-invariant while possessing an explicit regression logic. The method is performed on a large public remote sensing dataset DOTA and two popular detectors and achieves a large accuracy improvement in both large aspect ratio targets and small targets detection compared to similar methods.

ViT-UperNet: a hybrid vision transformer with unified-perceptual-parsing network for medical image segmentation

The existing image semantic segmentation models have low accuracy in detecting tiny targets or multi-targets at overlapping regions. This work proposes a hybrid vision transformer with unified-perceptual-parsing network (ViT-UperNet) for medical image segmentation. A self-attention mechanism is embedded in a vision transformer to extract multi-level features. The image features are extracted hierarchically from low to high dimensions using 4 groups of Transformer blocks with different numbers. Then, it uses a unified-perceptual-parsing network based on a feature pyramid network (FPN) and a pyramid pooling module (PPM) for the fusion of multi-scale contextual features and semantic segmentation. FPN can naturally use hierarchical features, and generate strong semantic information on all scales. PPM can better use the global prior knowledge to understand complex scenes, and extract features with global context information to improve segmentation results. In the training process, a scalable self-supervised learner named masked autoencoder is used for pre-training, which strengthens the visual representation ability and improves the efficiency of the feature learning. Experiments are conducted on cardiac magnetic resonance image segmentation where the left and right atrium and ventricle are selected for segmentation. The pixels accuracy is 93.85%, the Dice coefficient is 92.61% and Hausdorff distance is 11.16, which are improved compared with the other methods. The results show the superiority of Vit-UperNet in medical images segmentation, especially for the low-recognition and serious-occlusion targets.

Metal parçaların yüzey kusurlarını tespit için k-dedektörü ile küçük nesne tespit yöntemi

In the process of its development, intelligent manufacturing often focuses on production flexibility, client customization, and quality control, all of which are required for the manufacture of powder-based metallurgy. In particular, the identification and categorization of defects are crucial steps in the production processes involved in powder-based metallurgy. Intelligent strategies to detect faults in metal parts continue to be a challenge in automated industrial production lines. These techniques have been a particular concern for microscopic metal component producers for a long time. Due to its precision and speed, the YOLOv4 approach has been widely used for object detection. On the other hand, the identification of tiny targets, particularly imperfections on the surface of metal parts, continues to present a number of obstacles and difficulties. To increase the overall performance of detection, this research provided a technique for the detection of tiny objects based on YOLOv4 for such objects. To increase the effectiveness of the detection process, this involves expanding the size of the k detector while simultaneously eliminating unnecessary branches of the YOLO head network. Experiments have shown that the KD-YOLO model performs better than its predecessors, YOLOv4, YOLOv5, and PP-YOLO, in terms of the total number of parameters, classification accuracy and detection precision.

HDTNet: PCB defect detection algorithm for tiny objects of high-density regions

PCB defect detection aims to identify the presence of gaps, open circuits, short circuits, and other defects in the PCB boards produced in the industry. Designing effective deep learning algorithms is crucial to finding a solution. Previously proposed PCB defect detection algorithms are limited in detecting tiny objects in high-density. Directly applying previous models to tackle PCB defect detection tasks will cause serious issues, such as missed detection and false detection. In this paper, we present a detection algorithm for tiny PCB defect targets in high-density regions to solve the above-mentioned problems. We firstly propose a detection head to detect tiny objects. Then, we design a four-channel feature fusion mechanism to fuse four different scale features and add an attention mechanism to find the attention region in scenarios with dense objects. Finally, we achieved accurate detection of tiny targets in high-density areas. Experiments were performed on the publicly available PCB defect dataset from Peking University. Our mAP@.5:.95 achieves 48.6%, while mAP@0.5 exceeds 90%. Compared with YOLOX and YOLOv5, our improved model can better localize tiny objects in high-density scenes. The experimental results certify that our model can obtain higher performance in comparison with the baseline and the state of the art.

An efficient detection model based on improved YOLOv5s for abnormal surface features of fish.

Detecting abnormal surface features is an important method for identifying abnormal fish. However, existing methods face challenges in excessive subjectivity, limited accuracy, and poor real-time performance. To solve these challenges, a real-time and accurate detection model of abnormal surface features of in-water fish is proposed, based on improved YOLOv5s. The specific enhancements include: 1) We optimize the complete intersection over union and non-maximum suppression through the normalized Gaussian Wasserstein distance metric to improve the model's ability to detect tiny targets. 2) We design the DenseOne module to enhance the reusability of abnormal surface features, and introduce MobileViTv2 to improve detection speed, which are integrated into the feature extraction network. 3) According to the ACmix principle, we fuse the omni-dimensional dynamic convolution and convolutional block attention module to solve the challenge of extracting deep features within complex backgrounds. We carried out comparative experiments on 160 validation sets of in-water abnormal fish, achieving precision, recall, mAP50, mAP50:95 and frames per second (FPS) of 99.5, 99.1, 99.1, 73.9% and 88 FPS, respectively. The results of our model surpass the baseline by 1.4, 1.2, 3.2, 8.2% and 1 FPS. Moreover, the improved model outperforms other state-of-the-art models regarding comprehensive evaluation indexes.

YOLOv5s-MC: Lightweight road target detection network

For the problem of large number of target detection algorithm parameters, a lightweight real-time detection algorithm YOLOv5s-MC based on improved YOLOv5s road scenes is proposed. firstly, CA attention is added to the model to improve the sensitivity of the network to detect targets; secondly, in the feature fusion network, add adaptive weight parameters using AS-Concat structure are added to better fuse the feature information of different layers and improve the detection accuracy of the algorithm ; adding a small target detection layer to improve the detection accuracy of tiny targets; finally introducing Mobilnetv2, a lightweight network, as the overall backbone layer to realize the lightweight requirement of the network; to verify the advantages of the proposed algorithm, experiments were conducted on the kitti dataset. The experimental results show that the proposed algorithm, compared with the original network, improves the average accuracy by 0.2% with 55.8% less parameters and 33.7% less computation, and the detection speed reaches 35 FPS, which meets the requirements of real-time detection and improves the ability of algorithm deployment in weak hardware computing power scenarios to a certain extent.