Improve Detection Speed Research Articles

Vegetable disease detection using an improved YOLOv8 algorithm in the greenhouse plant environment

This study introduces YOLOv8n-vegetable, a model designed to address challenges related to imprecise detection of vegetable diseases in greenhouse plant environment using existing network models. The model incorporates several improvements and optimizations to enhance its effectiveness. Firstly, a novel C2fGhost module replaces partial C2f. with GhostConv based on Ghost lightweight convolution, reducing the model’s parameters and improving detection performance. Second, the Occlusion Perception Attention Module (OAM) is integrated into the Neck section to better preserve feature information after fusion, enhancing vegetable disease detection in greenhouse settings. To address challenges associated with detecting small-sized objects and the depletion of semantic knowledge due to varying scales, an additional layer for detecting small-sized objects is included. This layer improves the amalgamation of extensive and basic semantic knowledge, thereby enhancing overall detection accuracy. Finally, the HIoU boundary loss function is introduced, leading to improved convergence speed and regression accuracy. These improvement strategies were validated through experiments using a self-built vegetable disease detection dataset in a greenhouse environment. Multiple experimental comparisons have demonstrated the model's effectiveness, achieving the objectives of improving detection speed while maintaining accuracy and real-time detection capability. According to experimental findings, the enhanced model exhibited a 6.46% rise in mean average precision (mAP) over the original model on the self-built vegetable disease detection dataset under greenhouse conditions. Additionally, the parameter quantity and model size decreased by 0.16G and 0.21 MB, respectively. The proposed model demonstrates significant advancements over the original algorithm and exhibits strong competitiveness when compared with other advanced object detection models. The lightweight and fast detection of vegetable diseases offered by the proposed model presents promising applications in vegetable disease detection tasks.

A real-time and accurate convolutional neural network for fabric defect detection

As a practical and challenging task, deep learning-based methods have achieved effective results for fabric defect detection, however, most of them mainly target detection accuracy at the expense of detection speed. Therefore, we propose a fabric defect detection method called PEI-YOLOv5. First, Particle Depthwise Convolution (PDConv) is proposed to extract spatial features more efficiently while reducing redundant computations and memory access, reducing model computation and improving detection speed. Second, Enhance-BiFPN(EB) is proposed based on the structure of BiFPN to enhance the attention of spatial and channel feature maps and the fusion of information at different scales. Third, we improve the loss function and propose IN loss, which improves the problem that the original IOU loss is weak in detecting small targets while speeding up the convergence of the model. Finally, five more common types of defects were selected for training in the GuangDong TianChi fabric defect dataset, and using our proposed PEI-YOLOv5 with only 0.2 Giga Floating Point Operations (GFLOPs) increase, the mAP improved by 3.61%, reaching 87.89%. To demonstrate the versatility of PEI-YOLOv5, we additionally evaluated this in the NEU surface defect database, with the mAP of 79.37%. The performance of PEI-YOLOv 5 in these two datasets surpasses the most advanced fabric defect detection methods at present. We deployed the model to the NVIDIA Jetson TX2 embedded development board, and the detection speed reached 31 frames per second (Fps), which can fully meet the speed requirements of real-time detection.

Unleashing the power of AI in detecting metal surface defects: an optimized YOLOv7-tiny model approach.

The detection of surface defects on metal products during the production process is crucial for ensuring high-quality products. These defects also lead to significant losses in the high-tech industry. To address the issues of slow detection speed and low accuracy in traditional metal surface defect detection, an improved algorithm based on the YOLOv7-tiny model is proposed. Firstly, to enhance the feature extraction and fusion capabilities of the model, the depth aware convolution module (DAC) is introduced to replace all ELAN-T modules in the network. Secondly, the AWFP-Add module is added after the Concat module in the network's Head section to strengthen the network's ability to adaptively distinguish the importance of different features. Finally, in order to expedite model convergence and alleviate the problem of imbalanced positive and negative samples in the study, a new loss function called Focal-SIoU is used to replace the original model's CIoU loss function. To validate the effectiveness of the proposed model, two industrial metal surface defect datasets, GC10-DET and NEU-DET, were employed in our experiments. Experimental results demonstrate that the improved algorithm achieved detection frame rates exceeding 100 fps on both datasets. Furthermore, the enhanced model achieved an mAP of 81% on the GC10-DET dataset and 80.1% on the NEU-DET dataset. Compared to the original YOLOv7-tiny algorithm, this represents an increase in mAP of nearly 11% and 9.2%, respectively. Moreover, when compared to other novel algorithms, our improved model demonstrated enhanced detection accuracy and significantly improved detection speed. These results collectively indicate that our proposed enhanced model effectively fulfills the industry's demand for rapid and efficient detection and recognition of metal surface defects.

Study on the Lightweighting of YOLOv5s Model for Precise Detection of Irregular-shaped Components

Traditional methods for target recognition face challenges in meeting both precision and speed requirements for precision-shaped parts. In this paper, we propose an enhanced algorithm for precision-shaped part recognition by integrating deep learning theory. To achieve this, we modify the YOLOv5 network. Specifically, we replace the C3 module of the original network's backbone with the C3_ghostnetv2 network, which incorporates Ghostnetv2. This modification results in a lighter network with reduced model parameters and size, thereby improving detection speed. Moreover, we replace the convolution in the original network's neck with Deformable Convolution v2 (DCNv2) to enhance feature extraction for precision-shaped parts. We conduct comparative experiments on a self-made dataset of precision-shaped parts. The experimental results demonstrate that our improved algorithm reduces parameters by 13.7% and model size by 12.5% compared to the original YOLOv5s algorithm, while achieving a 1.4% increase in detection accuracy. The proposed algorithm accurately identifies and classifies precision-shaped machining parts, providing valuable technical support for subsequent intelligent production.

YOLOv3-MSSA based hot spot defect detection for photovoltaic power stations

With the continuous development of the energy industry, photovoltaic power generation is gradually becoming one of the main power generation methods. However, detecting hot spot defects in photovoltaic power stations is challenging. Therefore, enhancing detection efficiency using information technology has become a crucial aspect. The study presents a defect detection model for PV power stations using the YOLOv3 (You Only Look Once v3) algorithm. The model incorporates coordinate attention module (CAM) and self-attention module (SAM) to improve feature extraction in low-resolution conditions. The Multi objective Sparrow is employed to achieve multiple objectives. It is very contributing in the detection of low-resolution features. It shows that the research method can reduce the loss value to 0.009 after 400 iterations of the loss curve test. The precision-recall (P-R) curve generated by the research method only starts to drop sharply when the Recall value reaches 0.96. The number of parameters generated by the research method is 3.46×106. The detection accuracy of the research method reaches 98.86 % when there are five defective fault types. The results indicate that the proposed research method offers improved detection speed and higher accuracy in identifying hot spot defects in PV power stations. This technology provides valuable support for hot spot defect detection and presents new opportunities for the field.

Optimization study of anomaly detection algorithm in machine vision inspection technology

Abstract In recent years, target detection algorithms based on machine vision have been a hotspot in computer vision research. The You Only Look Once (YOLO) algorithm, as an excellent target detection algorithm, has played an important role in improving detection speed and accuracy with the improvement of the network architecture in its development process. This paper introduces the concept of integrated learning to the YOLOv5 network architecture, incorporating deformable convolution and attention mechanisms. It also chooses the focal EIOU loss function to replace the GIOU loss function, thereby addressing the issue of localization loss, prioritizing abnormally detected targets, and enhancing the detection efficiency of these abnormal targets. Finally, we examine the practical value of the improved YOLOv5 algorithm by testing its performance and applying it to real-world anomaly detection. The results show that the improved YOLOv5 model outperforms the original YOLOv5 model in terms of performance and practical application advantages. In terms of performance, the classification accuracy of sea_person and earth_person in the improved YOLOv5 model is 37% and 26%, respectively, which is a significant performance gain overall. In actual application tests, the proposed method is more accurate than the traditional method. The accuracy is significantly higher.

Zoom Text Detector.

To pursue comprehensive performance, recent text detectors improve detection speed at the expense of accuracy. They adopt shrink-mask-based text representation strategies, which leads to a high dependence of detection accuracy on shrink-masks. Unfortunately, three disadvantages cause unreliable shrink-masks. Specifically, these methods try to strengthen the discrimination of shrink-masks from the background by semantic information. However, the feature defocusing phenomenon that coarse layers are optimized by fine-grained objectives limits the extraction of semantic features. Meanwhile, since both shrink-masks and the margins belong to texts, the detail loss phenomenon that the margins are ignored hinders the distinguishment of shrink-masks from the margins, which causes ambiguous shrink-mask edges. Moreover, false-positive samples enjoy similar visual features with shrink-masks. They aggravate the decline of shrink-masks recognition. To avoid the above problems, we propose a zoom text detector (ZTD) inspired by the zoom process of the camera. Specifically, zoomed-out view module (ZOM) is introduced to provide coarse-grained optimization objectives for coarse layers to avoid feature defocusing. Meanwhile, zoomed-in view module (ZIM) is presented to enhance the margins recognition to prevent detail loss. Furthermore, sequential-visual discriminator (SVD) is designed to suppress false-positive samples by sequential and visual features. Experiments verify the superior comprehensive performance of ZTD.

FLLF: A Fast-Lightweight Location Detection Framework for False Data Injection Attacks in Smart Grids

This paper proposes a fast-lightweight location detection framework (FLLF) for false data injection attacks (FDIAs). The location detection of false data injection attacks is traditionally realized by computationally intensive neural networks, which can cause large detection delays due to the limited computation power and storage resources of the detection device. By contrast, the proposed method is lightweight and can be deployed on low-cost devices such as embedded devices, significantly improving detection speed while reducing power consumption. The proposed method consists of an efficient FDIA location detection model and an automatic model search algorithm. The model is a lightweight model that enables fast and accurate attack location detection through a combination of the early exiting mechanism and mixed-precision quantization (EE-MPQ). Then, an automatic model search algorithm is designed to enable the EE-MPQ model to match the smart grids with different structures. Finally, it is shown that the proposed method can accurately detect and locate FDIAs through numerical analysis on the IEEE 14-bus and 118-bus power systems.

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.

Research on rapid detection of cross-scale defects in surface based on deep learning

The complex and diverse forms of surface defects in metal cutting, as well as their large scale span, present new challenges for deep learning algorithms. In addition, the existing defect detection models are generally characterized by high computational amount and complex structures, which is contrary to the high real-time performance and limited computing resources required in industrial applications. Based on this, this paper proposes a rapid detection method for cross-scale defects in surfaces based on deep learning. Firstly, a dataset of defect surfaces is collected and constructed through cutting experiments. Then, the experimental analysis reveals the insufficiency of the You Only Look Once version-5 s (YOLOv5s) network model for the detection of cross-scale defects on the surface. As a result, a RepVGG-Coordinate Attention-YOLOv5s (Rep-CA-YOLOv5s) network model, suitable for cross-scale defect detection, is proposed. This model optimizes the YOLOv5s network model from three perspectives, enhancing its ability to extract and fuse features for cross-scale defects. Finally, this paper investigates methods to improve detection speed while ensuring model accuracy. Two optimal Rep-CA-YOLOv5s sparse models are obtained through sparse training based on the γ scaling factor of the Batch Normalization (BN) layer and the filter weight, respectively. The relationship between detection accuracy, parameter quantity, and inference speed of these two sparse models under different pruning rates is explored. Experimental results indicate that the filter pruning method significantly improves model inference speed. At a 50 % pruning rate, optimal detection results can be achieved. Compared with the unpruning model, the pruned model reduced the inference speed by 55.67 %, while the mean Average Precision (mAP) only decreased by 0.1 %.

GSD-YOLOX: Lightweight and more accurate object detection models

We present an enhanced YOLOX model for vehicle detection, addressing issues of slow detection speed, high parameter counts, and complex computations. Our model significantly improves inference speed while maintaining high detection accuracy. We introduce two lightweight modules, DG and DS, to reduce model size and enhance detection speed. The DG module eliminates redundant features during feature extraction, and the DS module optimizes the performance of DG, enhancing feature extraction efficiency. We utilize the CIoU loss function for accurate bounding box regression. Additionally, we introduce the Chinese Chongqing Vehicle Detection Benchmark (CCVTDB) dataset to address dataset limitations. Our lightweight model achieves an impressive 83.36% detection accuracy on CCVTDB at 65 FPS with 4.4 million parameters. Compared to the original model, our approach improves detection speed by 30% and reduces model size by 51%, while maintaining substantial detection performance.

MobileOne-YOLO: Improving the YOLOv7 network for the detection of unfertilized duck eggs and early duck embryo development - a novel approach

Unfertilized duck eggs and early dead embryo eggs can significantly harm normally developing embryo eggs if not detected and removed in a timely manner. The traditional light-illumination technique currently used for this purpose is inefficient, time-consuming, and labor-intensive, making it unsuitable for modern farms and hatcheries. To address this issue, this study focuses on duck eggs incubated for a period of 10 days. This paper introduces a newly developed duck egg image acquisition device that incorporates a duck egg suction lifter with a built-in light source. The device is designed based on the incubation process and has the potential for rapid implementation in actual production. Furthermore, this study describes an improved algorithm for identifying unfertilized duck eggs and early dead embryo eggs based on the YOLOv7 network. The algorithm incorporates the use of MobileOne as a replacement for YOLOv7′s backbone network and lightweight operations applied to the YOLO head. The test results demonstrate that the MobileOne-YOLO achieves 98.10% detection accuracy, 98.55% recall, 97.79% mAP50, 98.55% precision, and 142.8 FPS in the test set images. In comparison, the original YOLOV7 also achieves 98.10% detection accuracy, 97.10% recall, 97.86% mAP50, 100.00% precision, and 101.2 FPS. The MobileOne-YOLO network maintains the same level of detection accuracy as the YOLOV7 network while achieving significantly improved detection speed. The technical method proposed in this study meets the actual production requirements in terms of detection accuracy and speed. Moreover, the image acquisition method adopted in this study is based on the incubation process, making it readily applicable to production practice. This study provides a new foundation for the subsequent development of intelligent detection equipment for identifying unfertilized duck eggs and early dead embryo eggs.

Strip steel surface defect detection based on lightweight YOLOv5.

Deep learning-based methods for detecting surface defects on strip steel have advanced detection capabilities, but there are still problems of target loss, false alarms, large computation, and imbalance between detection accuracy and detection speed. In order to achieve a good balance between detection accuracy and speed, a lightweight YOLOv5 strip steel surface defect detection algorithm based on YOLOv5s is proposed. Firstly, we introduce the efficient lightweight convolutional layer called GSConv. The Slim Neck, designed based on GSConv, replaces the original algorithm's neck, reducing the number of network parameters and improving detection speed. Secondly, we incorporate SimAM, a non-parametric attention mechanism, into the improved neck to enhance detection accuracy. Finally, we utilize the SIoU function as the regression prediction loss instead of the original CIoU to address the issue of slow convergence and improve efficiency. According to experimental findings, the YOLOv5-GSS algorithm outperforms the YOLOv5 method by 2.9% on the NEU-DET dataset and achieves an average accuracy (mAP) of 83.8% with a detection speed (FPS) of 100 Hz, which is 3.8 Hz quicker than the YOLOv5 algorithm. The proposed model outperforms existing approaches and is more useful, demonstrating the efficacy of the optimization strategy.

Research and Explainable Analysis of a Real-Time Passion Fruit Detection Model Based on FSOne-YOLOv7

Real-time object detection plays an indispensable role in facilitating the intelligent harvesting process of passion fruit. Accordingly, this paper proposes an FSOne-YOLOv7 model designed to facilitate the real-time detection of passion fruit. The model addresses the challenges arising from the diverse appearance characteristics of passion fruit in complex growth environments. An enhanced version of the YOLOv7 architecture serves as the foundation for the FSOne-YOLOv7 model, with ShuffleOne serving as the novel backbone network and slim-neck operating as the neck network. These architectural modifications significantly enhance the capabilities of feature extraction and fusion, thus leading to improved detection speed. By utilizing the explainable gradient-weighted class activation mapping technique, the output features of FSOne-YOLOv7 exhibit a higher level of concentration and precision in the detection of passion fruit compared to YOLOv7. As a result, the proposed model achieves more accurate, fast, and computationally efficient passion fruit detection. The experimental results demonstrate that FSOne-YOLOv7 outperforms the original YOLOv7, exhibiting a 4.6% increase in precision (P) and a 4.85% increase in mean average precision (mAP). Additionally, it reduces the parameter count by approximately 62.7% and enhances real-time detection speed by 35.7%. When compared to Faster-RCNN and SSD, the proposed model exhibits a 10% and 4.4% increase in mAP, respectively, while achieving approximately 2.6 times and 1.5 times faster real-time detection speeds, respectively. This model proves to be particularly suitable for scenarios characterized by limited memory and computing capabilities where high accuracy is crucial. Moreover, it serves as a valuable technical reference for passion fruit detection applications on mobile or embedded devices and offers insightful guidance for real-time detection research involving similar fruits.

PCBSNet: A Pure Convolutional Bilateral Segmentation Network for Real-Time Natural Scene Text Detection

Scene text detection is a fundamental research work in the field of image processing and has extensive application value. Segmentation-based methods have time-consuming feature processing, while post-processing algorithms are excellent. Real-time semantic segmentation methods use lightweight backbone networks for feature extraction and aggregation but lack effective post-processing methods. The pure convolutional network improves model performance by changing key components. Combining the advantages of three types of methods, we propose a Pure Convolutional Bilateral Segmentation Network (PCBSNet) for real-time natural scene text detection. First, we constructed a bilateral feature extraction backbone network to significantly improve detection speed. The low extraction detail branch captures spatial information, while the efficient semantic extraction branch accurately captures semantic features through a series of micro designs. Second, we built an efficient attention aggregation module to guide the efficient and adaptive aggregation of features from the two branches. The fused feature map undergoes feature enhancement to obtain more accurate and reliable feature representation. Finally, we used differentiable binarization post-processing to construct text instance boundaries. To evaluate the effectiveness of the proposed model, we compared it with mainstream lightweight models on three datasets: ICDAR2015, MSRA-TD500, and CTW1500. The F-measure scores were 82.9%, 82.8%, and 78.9%, respectively, and the FPS were 59.1, 94.3, and 75.5 frames per second. We also conducted extensive ablation experiments on the ICDAR2015 dataset to validate the rationality of the proposed improvements. The obtained results indicate that the proposed model significantly improves inference speed while enhancing accuracy and demonstrates good competitiveness compared to other advanced detection methods. However, when faced with curved text, the detection performance of PCBSNet needs to be improved.

Wireless Material Identification in the Recycling Chain Using Chipless RFID Tags

The utility of chipless RFID technology for the identification of 8 materials (7 plastic types and glass) in the recycling chain is examined. We first demonstrate how the frequency response of 8 Circular Ring Resonators (CRR), designed to operate in the 1.5 - 6 GHz band, is used to encode 8 unique material IDs. Tagged items are inserted one at a time in front of a chipless RFID reader and the speed and accuracy of material identification, based on the ability to decode the tag’s frequency response, is determined. We demonstrate the ability to correctly identify material type using a random forest classifier with up to 93% accuracy while being 3-16X faster than other methods reported in the literature. A technique to further improve detection speed, taking measurements with lower resolution, at the cost of accuracy is then presented. Finally, future directions of exploration are discussed.

Study on Lightweight Model of Maize Seedling Object Detection Based on YOLOv7

Traditional maize seedling detection mainly relies on manual observation and experience, which is time-consuming and prone to errors. With the rapid development of deep learning and object-detection technology, we propose a lightweight model LW-YOLOv7 to address the above issues. The new model can be deployed on mobile devices with limited memory and real-time detection of maize seedlings in the field. LW-YOLOv7 is based on YOLOv7 but incorporates GhostNet as the backbone network to reduce parameters. The Convolutional Block Attention Module (CBAM) enhances the network’s attention to the target region. In the head of the model, the Path Aggregation Network (PANet) is replaced with a Bi-Directional Feature Pyramid Network (BiFPN) to improve semantic and location information. The SIoU loss function is used during training to enhance bounding box regression speed and detection accuracy. Experimental results reveal that LW-YOLOv7 outperforms YOLOv7 in terms of accuracy and parameter reduction. Compared to other object-detection models like Faster RCNN, YOLOv3, YOLOv4, and YOLOv5l, LW-YOLOv7 demonstrates increased accuracy, reduced parameters, and improved detection speed. The results indicate that LW-YOLOv7 is suitable for real-time object detection of maize seedlings in field environments and provides a practical solution for efficiently counting the number of seedling maize plants.

Lightweight algorithm for multi-scale ship detection based on high-resolution SAR images

ABSTRACT As ship target detection technology has high application value in military and civil fields, it is significant to research ship detection in SAR images. Aiming at the complex and diverse backgrounds, significant differences in ship sizes, and real-time detection problems in the ship target detection task of SAR remote sensing images, a lightweight ship detection network based on the YOLOx-Tiny model is proposed. Firstly, a multi-scale ship feature extraction module is proposed, composed of a parallel multi-branch structure connected by a standard convolution layer, asymmetric convolution layer, and dilatation convolution layer with different expansion rates in turn. It makes better use of local features and global features and effectively improves the detection accuracy of multi-scale ship targets; Secondly, to ensure detection performance and eliminate background interference, we propose a whole SAR remote sensing image detection strategy based on an adaptive threshold, which effectively suppresses false alarms caused by background and improves detection speed. The experimental results on two different SAR ship datasets, SSDD and HRSID, show that, compared with several advanced methods, the effectiveness and superiority of the method in this paper are verified, and excellent results are shown in the detection of the whole SAR remote-sensing image. It can provide effective theoretical and technical support for ship detection on platforms with limited computing resources and has good application prospects.

Real-time detection method for welding parts completeness based on improved YOLOX in a digital twin environment

Detection of the completeness of welding parts for automotive body-in-white welding relies mainly on both artificial and sensor detection. Due to a lack of intelligent methods, it is difficult to achieve accurate detection. This paper presents a new intelligent detection method based on improved YOLOX in a digital twin (DT) environment. Firstly, to address the problem of insufficient real samples, virtual datasets are made to increase data volume by using DT technology and realize the fusion of virtuality and reality. Secondly, an improved MobileNetv1 network is designed as the feature extraction network for YOLOX. Additionally, the original convolution is replaced by depthwise separable convolution blocks for reducing computational burden and improving detection speed. Experimental results show that the number of parameters is 59.1% less than that of the original model and the detection speed is increased from 36 to 50 frames s–1. Meanwhile, mean average precision increases by 1.42% and 2.76%, respectively, under two different overlaps.

Underwater Forward-Looking Sonar Images Target Detection via Speckle Reduction and Scene Prior

Forward-looking sonar (FLS) imagery system plays a significant role in oceanic object recognition and detection since it can overcome the limitation of lighting conditions and reflect the real situation of the underwater environment. However, object detection algorithms for FLS images remain challenging for two main reasons: 1) The noise caused by the coherent characteristic of the scattering phenomenon impairs the detector capture of target information; 2) The scene prior based on the uneven target scale distribution is generally neglected, which leads to the detector generating redundant anchors and slows down detection efficiency. Confronting such challenges, this paper characterizes the noise and the uneven target scale distribution in FLS images as multiplicative speckle noise and scene prior, respectively. Therefore, we propose a novel underwater FLS image detection network, namely UFIDNet, to further improve detection performance by considering speckle noise reduction and scene prior in FLS images. More specifically, a speckle reduction auxiliary branch (SRAB) is designed to introduce additional despeckled supervision information to encourage the feature extractor to produce clean features and share them with the detection pipeline during the training phase. In particular, the noise distribution of FLS images is excavated for synthetic dataset construction and Despeckle Network (DSN) design to obtain despeckled supervision images. In addition, a feature selection strategy (FSS) embedded in detection branch is designed to screen out feature levels that do not match the target size, thus significantly reducing the generation of redundant anchors and improving detection speed. Experimental results show that our UFIDNet achieves 70.5% and 47.3% average precision (AP), 81.3% and 54.6% average recall (AR <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">max=10</sub> ), 27.0 and 26.1 FPS on two real FLS datasets respectively, outperforming many state-of-the-art general detectors and sonar image detectors.