Small Size Objects Research Articles

Domain Randomization Using Deep Neural Networks for Estimating Positions of Bolts

Abstract Current manual practices of replacing bolts on structures are time-consuming and costly, especially because of numerous bolts. Thus, an automated method that can visually detect and localize bolt positions would be highly beneficial. We demonstrate the use of deep neural networks using domain randomization for detecting and localizing bolts on a workpiece. In contrast to previous approaches that require training on real images, the use of domain randomization enables all training in simulation. The key idea is to create a wide variety of computer-generated synthetic images by varying the texture, color, camera position and orientation, distractor objects, and noise, and train the neural network on these images such that the neural network is robust to scene variability and hence provides accurate results when deployed on real images. Using domain randomization, we train two neural networks, a faster regional convolutional neural network for detecting the bolt and placing a bounding box, and a regression convolutional neural network for estimating the x- and y-position of the bolts relative to the coordinates fixed to the workpiece. Our results indicate that in the best case, we can detect bolts with 85% accuracy and can predict 75% of bolts within 1.27 cm accuracy. The novelty of this work is in using domain randomization to detect and localize: (1) multiples of a single object and (2) small-sized objects (0.6 cm × 2.5 cm).

Segmenting Purple Rapeseed Leaves in the Field from UAV RGB Imagery Using Deep Learning as an Auxiliary Means for Nitrogen Stress Detection

Crop leaf purpling is a common phenotypic change when plants are subject to some biotic and abiotic stresses during their growth. The extraction of purple leaves can monitor crop stresses as an apparent trait and meanwhile contributes to crop phenotype analysis, monitoring, and yield estimation. Due to the complexity of the field environment as well as differences in size, shape, texture, and color gradation among the leaves, purple leaf segmentation is difficult. In this study, we used a U-Net model for segmenting purple rapeseed leaves during the seedling stage based on unmanned aerial vehicle (UAV) RGB imagery at the pixel level. With the limited spatial resolution of rapeseed images acquired by UAV and small object size, the input patch size was carefully selected. Experiments showed that the U-Net model with the patch size of 256 × 256 pixels obtained better and more stable results with a F-measure of 90.29% and an Intersection of Union (IoU) of 82.41%. To further explore the influence of image spatial resolution, we evaluated the performance of the U-Net model with different image resolutions and patch sizes. The U-Net model performed better compared with four other commonly used image segmentation approaches comprising support vector machine, random forest, HSeg, and SegNet. Moreover, regression analysis was performed between the purple rapeseed leaf ratios and the measured N content. The negative exponential model had a coefficient of determination (R²) of 0.858, thereby explaining much of the rapeseed leaf purpling in this study. This purple leaf phenotype could be an auxiliary means for monitoring crop growth status so that crops could be managed in a timely and effective manner when nitrogen stress occurs. Results demonstrate that the U-Net model is a robust method for purple rapeseed leaf segmentation and that the accurate segmentation of purple leaves provides a new method for crop nitrogen stress monitoring.

Refining the Uluzzian through a new lithic assemblage from Roccia San Sebastiano (Mondragone, southern Italy)

Roccia San Sebastiano is a tectonic-karstic cave located at the foot of the southern slope of Mt. Massico, in the territory of Mondragone (Caserta) in Campania (southern Italy). Systematic excavation has been carried out since 2001, leading to the partial exploration of an important Pleistocene deposit, extraordinarily rich in lithic and faunal remains. The aim of this paper is to (1) present the stratigraphic sequence of Roccia San Sebastiano, and (2) technologically describe the lithic materials of squares F14 t18, t19, t20; E16 t16, t17, t18 recently recognised as Uluzzian. The stratigraphic sequence is more than 3 m thick and dates from the Middle to the Upper Palaeolithic. It contains different techno-complexes: Gravettian, Aurignacian, Uluzzian and Mousterian. In the Uluzzian lithic assemblage mostly local pebbles of chert were used in order to produce small-sized objects. The concept of debitage mainly deals with unidirectional debitage with absent or fairly accurate management of the convexities and angles; the striking platforms are usually natural or made by one stroke. It is attested the use of both direct freehand percussion and bipolar technique on anvil in the same reduction sequence. Amongst the retouched tools the presence of two lunates is of note. This study of the Roccia San Sebastiano Uluzzian lithic complexes is significant for understanding the dynamics of the transition from Middle to Upper Palaeolithic in the Tyrrhenian margin of southern Italy.

Object tracking algorithm for a passive positioning system

We propose an algorithm for small-sized mobile object detection and trajectory parameter estimation for a passive positioning system that consists of several optical, thermal, and radio sensors. The algorithm is based on a combination of spatial and temporal processing of observation data. For spatial processing, a set of equations is solved that defines the sufficient condition for coupling the direction vectors to probable objects in the image stereo pair. Object coordinates and velocities for a single observation period are estimated. For temporal processing, the direction vectors are distributed based on connection to probable objects in a sequence of the capture intervals. The results of numerical modeling of the proposed algorithm show the advantage of combining the two approaches in comparison with the traditional object detection and tracking algorithms.

Applications of Deep Learning for Dense Scenes Analysis in Agriculture: A Review.

Deep Learning (DL) is the state-of-the-art machine learning technology, which shows superior performance in computer vision, bioinformatics, natural language processing, and other areas. Especially as a modern image processing technology, DL has been successfully applied in various tasks, such as object detection, semantic segmentation, and scene analysis. However, with the increase of dense scenes in reality, due to severe occlusions, and small size of objects, the analysis of dense scenes becomes particularly challenging. To overcome these problems, DL recently has been increasingly applied to dense scenes and has begun to be used in dense agricultural scenes. The purpose of this review is to explore the applications of DL for dense scenes analysis in agriculture. In order to better elaborate the topic, we first describe the types of dense scenes in agriculture, as well as the challenges. Next, we introduce various popular deep neural networks used in these dense scenes. Then, the applications of these structures in various agricultural tasks are comprehensively introduced in this review, including recognition and classification, detection, counting and yield estimation. Finally, the surveyed DL applications, limitations and the future work for analysis of dense images in agriculture are summarized.

Real-time traffic sign detection and classification towards real traffic scene

In this paper we propose a real-time traffic sign recognition algorithm which is robust to the small-sized objects and can identify all traffic sign categories. Specifically, we present a two-level detection framework which consists of the region proposal module(RPM) which is responsible for locating the objects and the classification module(CM) which aims to classify the located objects. In addition, to solve the problem of insufficient samples, we present an effective data augmentation method based on traffic sign logo to generate enough training data. The experiments are conducted in TT100k, and the results show the superiority of our method.

Robust Vehicle Detection on Multi- Resolution Aerial Images

Detecting vehicles in aerial images is one of the core components for intelligent transportation system. This task is challenging due to the comparably small size of the target objects, the complex background and multi-perspective views. It is particularly difficult for the real-time detection cases where only several tens of milliseconds delays are allowed. In this paper, we propose an approach to detect vehicles from aerial images with different resolutions and perspectives in an approximate real-time manner. The proposed model is robust and can detect vehicles in various detection scenarios. It can detect vehicles from an image within 47 milliseconds. We evaluate our method on a challenging data set of original aerial images over Munich and our data set collected using an unmanned aerial vehicle (UAV). The experimental results have demonstrated that our proposed method is superior to the state-of-the- art algorithms with respect to accuracy, recall and detection time.

Mixed Road User Trajectory Extraction From Moving Aerial Videos Based on Convolution Neural Network Detection

Vehicle trajectory data under mixed traffic conditions provides critical information for urban traffic flow modeling and analysis. Recently, the application of unmanned aerial vehicles (UAV) creates a potential of reducing traffic video collection cost and enhances flexibility at the spatial-temporal coverage, supporting trajectory extraction in diverse environments. However, accurate vehicle detection is a challenge due to facts such as small vehicle size and inconspicuous object features in UAV videos. In addition, camera motion in UAV videos hardens the trajectory construction procedure. This research aims at proposing a novel framework for accurate vehicle trajectory construction from UAV videos under mixed traffic conditions. Firstly, a Convolution Neural Network (CNN)-based detection algorithm, named You Only Look Once (YOLO) v3, is applied to detect vehicles globally. Then an image registration method based on Shi-Tomasi corner detection is applied for camera motion compensation. Trajectory construction methods are proposed to obtain accurate vehicle trajectories based on data correlation and trajectory compensation. At last, the ensemble empirical mode decomposition (EEMD) is applied for trajectory data denoising. Our framework is tested on three aerial videos taken by an UAV on urban roads with one including intersection. The extracted vehicle trajectories are compared with manual counts. The results show that the proposed framework achieves an average Recall of 91.91% for motor vehicles, 81.98% for non-motorized vehicles and 78.13% for pedestrians in three videos.

An Optimized Deep Neural Network Detecting Small and Narrow Rectangular Objects in Google Earth Images

Object detection is an important task for rapidly localizing target objects using high-resolution satellite imagery (HRSI). Although deep learning has been shown an efficient means of detection, object detection in HRSI remains problematic due to variations in object scale and size. In this article, we present a novel deep neural network (DNN) that combines double-shot neural network with misplaced localization strategy that adapts to object detection tasks in satellite images. This novel architecture optimizes the localization of small and narrow rectangular objects, which frequently appear in HRSI images, without accuracy loss on other size and width/height ratio objects. This method outperforms other state-of-art methods. We evaluated our proposed method on the NWPU VHR-10 public dataset and a new benchmark dataset (seven classes of small and narrow rectangular objects, SNRO-7). The NWPU VHR-10 dataset built a dataset for multiclass object detection; however, most labels are assigned in normal size and width/height ratios. SNRO-7 focuses on multiscale and multisize object detection and includes many small-size and narrow rectangular objects. We also evaluated the accuracy difference on DNN training and testing between gray scale and RGB datasets. The results of the experiment on object detection reveal that the mean average precision (MaP) of our method is 82.6% in NWPU VHR-10 and 79.3% in SNRO-7, which exceeds the MaPs of other state-of-the-art object detection neural networks. The model trained with the RGB dataset can achieve similar accuracy (around 79.0% MIoU) testing in both RGB and gray scale datasets. When training the model by mixing RGB and gray scale datasets in different ratios, the accuracy in the RGB channel significantly decreases with increasing gray scale images, but this does not influence the accuracy in the gray scale dataset.

ПРОЕКЦІЙНО-РАСТРОВІ МЕТОДИ РЕЄСТРАЦІЇ ПЕРЕТИНІВ ПОВЕРХНІ ЛОКАЛЬНИХ ОБ'ЄКТІВ

In work is solved the problem of visualizing terrain contour lines on the surface of the volume bodies. Modern methods andmeans of digital photogrammetry represent a real opportunity to return to the previously known analog methods of visualization,real-time monitoring, analysis and evaluation of the surface shape of the local volume bodies. the Essence of these methodsconsists in designing a linear raster digital projector or laser light plane on the object and the real-time monitoring and, ifnecessary, photographing localized on the family member contour and processing using modern software and hardware anddigital technology. This technique was used quite successfully in the study of local objects of small size in many applicationareas of applied photogrammetry: architecture and construction, studying of monuments of historical and cultural heritage, ship,air, auto, construction, medicine, etc. the paper presents circuits and analysis of their application with the shooting design oflinear raster parallel beam by moving the light plane, the Central design of the raster, the mathematical apparatus of calculationof parameters of acquisition and the results of the survey of the surface of the cone, the blades, and sculptures. The generalizedformulas allow you to select the shooting options of objects to perform calculations and get the dimensions and a graphicalmodel of the contour surfaces.This article is intended for students and teachers in the field of knowledge Architecture and engineering and professionalsworking in the field of application of modern methods of applied photogrammetry.

ЧИСЛЕННОЕ МОДЕЛИРОВАНИЕ ИСПЫТАНИЯ ПАДАЮЩИМ ГРУЗОМ ВЯЗКОЙ ТРУБНОЙ СТАЛИ ПРИ ТРЕХТОЧЕЧНОМ ИЗГИБЕ

The processes in the metal sample of a supply pipeline realized under drop-weight tests (DWT, or DWTT according to ASTM) are studied. DWT is a proof test of the pipeline metal that should ensure high resistance of the pipeline against extensive destruction. Numerical simulation of DWT with the steel sample of full thickness was performed; the steel had К65 strength grade. Parallel finite-element computer code DANCO developed in RFNC-VNIIEF was used for simulations. A detailed description of the rupture formation process required a fine-enough mesh and a supercomputer. To carry out the numerical simulation of the process, the constants of the deformation diagram were used, obtained on the basis of static and dynamic tensile tests of samples at room temperature. A modified Gurson–Tvergaard–Niedelman (GTNm) model for macro-viscous steel destruction (ductile failure) was used to describe the strain and the destruction of the metal. The modification makes it possible to describe direct and oblique cuts and their combinations in case of ductile failure of small-size objects (rods, plates, shells). The calculated dependences of the movement of the crack tip on the movement of the load and the resistance force of the sample on the movement of the crack tip are presented. We have got a good agreement between the computations and experimental data with regard to the “force–displacement” strength parameter, the deformed profiles and macro-geometry of the ruptured sample after the tests. The computation results reveal the mechanics of the crack origination, start and propagation in the sample, describe the plastic-flow energy distribution in the process of dynamic destruction. The results of the work can be used in the development of the requirements and of the implementation conditions of the “tooled” DWT, and for numerical simulation of the extensive destruction at the main pipeline.

The background noise effect on achieved sensitivity of heat source direction finding channels of optoelectronic systems used for detecting small-sized objects against the background of sky and sea

The background noise effect on achieved sensitivity of heat source direction finding channels of optoelectronic systems used for detecting small-sized objects against the background of sky and sea

Real-Time Detection and Recognition of Multiple Moving Objects for Aerial Surveillance

Detection of moving objects by unmanned aerial vehicles (UAVs) is an important application in the aerial transportation system. However, there are many problems to be handled such as high-frequency jitter from UAVs, small size objects, low-quality images, computation time reduction, and detection correctness. This paper considers the problem of the detection and recognition of moving objects in a sequence of images captured from a UAV. A new and efficient technique is proposed to achieve the above objective in real time and in real environment. First, the feature points between two successive frames are found for estimating the camera movement to stabilize sequence of images. Then, region of interest (ROI) of the objects are detected as the moving object candidate (foreground). Furthermore, static and dynamic objects are classified based on the most motion vectors that occur in the foreground and background. Based on the experiment results, the proposed method achieves a precision rate of 94% and the computation time of 47.08 frames per second (fps). In comparison to other methods, the performance of the proposed method surpasses those of existing methods.

Multiple-Oriented and Small Object Detection with Convolutional Neural Networks for Aerial Image

Detecting objects in aerial images is a challenging task due to multiple orientations and relatively small size of the objects. Although many traditional detection models have demonstrated an acceptable performance by using the imagery pyramid and multiple templates in a sliding-window manner, such techniques are inefficient and costly. Recently, convolutional neural networks (CNNs) have successfully been used for object detection, and they have demonstrated considerably superior performance than that of traditional detection methods; however, this success has not been expanded to aerial images. To overcome such problems, we propose a detection model based on two CNNs. One of the CNNs is designed to propose many object-like regions that are generated from the feature maps of multi scales and hierarchies with the orientation information. Based on such a design, the positioning of small size objects becomes more accurate, and the generated regions with orientation information are more suitable for the objects arranged with arbitrary orientations. Furthermore, another CNN is designed for object recognition; it first extracts the features of each generated region and subsequently makes the final decisions. The results of the extensive experiments performed on the vehicle detection in aerial imagery (VEDAI) and overhead imagery research data set (OIRDS) datasets indicate that the proposed model performs well in terms of not only the detection accuracy but also the detection speed.

Nonlinear representation of the confidence region of orbits determined on short arcs

The total number of active satellites, rocket bodies, and debris larger than 10 cm is currently about 20,000. If all resident space objects larger than 1 cm are considered, this number increases to an estimate of 700,000 objects. The next generation of sensors will be able to detect small-size objects, producing millions of observations per day. However, due to observability constraints, long gaps between observations will be likely to occur, especially for small objects. As a consequence, when acquiring observations on a single arc, an accurate determination of the space object orbit and the associated uncertainty is required. This work aims to revisit the classical least squares method by studying the effect of nonlinearities in the mapping between observations and state. For this purpose, high-order Taylor expansions enabled by differential algebra are exploited. In particular, an arbitrary-order least squares solver is implemented using the high-order expansion of the residuals with respect to the state. Typical approximations of differential correction methods are then avoided. Finally, the confidence region of the solution is accurately characterized with a nonlinear approach, taking advantage of these expansions. The properties and performance of the proposed methods are assessed using optical observations of objects in LEO, HEO, and GEO.

Biochemical sensor simulation for rectangular metal nanoparticles

To enhance sensor sensitivity and resolution, the present study proposes a biochemical sensor utilizing localized surface plasmon resonance to detect minute changes in the refractive index of the analyte. This sensor comprises a single-mode fiber located on the front and back ends of the sensor, which is responsible for the input and output of fiber optic signals, and two geometrical objects arranged in an interwoven pattern in the sensing region at the center of the sensor. In particular, the rectangular metal nanoparticles are arranged in a rectangular fashion and plated in the etched cladding layer to increase the number of nanosized metal particles on the X–Y plane. In addition, to decrease the calculation time and memory capacity while increasing the precision of simulations, a numerical calculation approach with power functions was employed. This approach comprises the finite element method, eigenmode expansion method, perfectly matched layer, perfectly reflecting boundary condition, boundary meshing method, and object meshing method. Triangular meshes were employed as the basis and sensors were divided into object boundaries, small-sized objects, medium-sized objects, and large-sized objects. The meshing ratio was set to 1:9:181:1810. Numerical simulation results indicate that the proposed localized surface plasmon resonance sensor achieved superior length (approx. 170 μm), resolution (approx. −155 dB), and sensitivity (approx. 137,664.4 nm/RIU).

Dual-laser-actuated operation of small size objects at a liquid interface.

This work focuses on the use, for the first time to our knowledge, of dual laser beams in photothermal-effect-based propulsion of small size objects at liquid interfaces. Compared with the single-laser mode, dual-laser-actuated operation turns out to be much more controllable with high quality, efficiency, and anti-interference capacity, which can be achieved through automated programming instead of through manual operation. A series of experiments were carried out to verify the principle, with the effects of laser power, laser-spot distance, and movement speed discussed in detail. The findings of this work might provide some insights into the development of intelligent macro/micro-operation systems for manipulating objects at different scales, such as drug particles and cells at liquid interfaces in the future.

IDENTIFICATION AND DETERMINATION OF LIGHT MARKS WHEN THEY OVERLAP IN OPTICAL CONTROL SYSTEMS

The issues of identification and determination of the relative position of localized light marks and images of small-sized objects recorded by an optical-electronic system (OES) are considered. A physical model and a research virtual device (VI) were developed on the basis of National Instruments software for identification and coordinate determination of two light marks at their different relative positions and the ratio of their intensities. It is shown that the use of a mathematical method for analyzing the illumination contour of two light labels on the basis of a continuous wavelet transform (CWT) allows identifying each label separately and determining the degree of superposition of their images. It is shown that the amplitudes and positions of the maxima of the CWT coefficients are related to the relative shift of the illuminance contours of the light marks, as the marking contours separate, the dependence of the position of the maxima becomes linear. The result of comparing the experimental data with theoretical calculations for the models of light labels showed a match within 25 %.

Development of mathematical model of localization of a small explosive object with the help of a specialized protective device

In the work, as a research object, a protective device of a domed form is used which is used by pyrotechnic units to localize an emergency situation in the event of an explosion inside a small hazardous object. It is noted that one of the most problematic places of its application is the development of recommendations, implementation of which should ensure the prevention of the development of an emergency event up to a level of emergency on such priority effects as the number of victims and the number of victims. It is shown that the definition of such recommendations, providing localization of the consequences of an emergency in the case of an explosion of a small explosive object inside a specialized protective device, requires the obtaining of a mathematical model of localization of the focal point of an emergency. This model should provide an assessment of the strength of the technical means of localization of fragments and become the basis for the correction of existing operational procedures in the case of its application by pyrotechnic units. In the course of the study, the Eulerian-Lagrangian approach is used, which would allow obtaining a mathematical model of localization with the help of a dome-shaped form of the consequences of emergency situations in the event of an explosion inside a small-sized dangerous object. In practice, a mathematical model is implemented in a finite element packet using the library of the ANSYS/AUTODYN computer system. This allows not to create an actual new package of applications every time, as was done before to describe similar models. Due to this, an assessment of the strength of the technical means of localization of fragments is provided. In comparison with similar well-known models, the developed mathematical model allows to estimate the size of the minimum thickness of the protective device. It allows to withstand the explosion of a small-sized explosive object and to determine the minimum mass of the protective equipment taking into account the operational capabilities of the combat calculation of the pyrotechnic unit.

ACF Based Region Proposal Extraction for YOLOv3 Network Towards High-Performance Cyclist Detection in High Resolution Images.

You Only Look Once (YOLO) deep network can detect objects quickly with high precision and has been successfully applied in many detection problems. The main shortcoming of YOLO network is that YOLO network usually cannot achieve high precision when dealing with small-size object detection in high resolution images. To overcome this problem, we propose an effective region proposal extraction method for YOLO network to constitute an entire detection structure named ACF-PR-YOLO, and take the cyclist detection problem to show our methods. Instead of directly using the generated region proposals for classification or regression like most region proposal methods do, we generate large-size potential regions containing objects for the following deep network. The proposed ACF-PR-YOLO structure includes three main parts. Firstly, a region proposal extraction method based on aggregated channel feature (ACF) is proposed, called ACF based region proposal (ACF-PR) method. In ACF-PR, ACF is firstly utilized to fast extract candidates and then a bounding boxes merging and extending method is designed to merge the bounding boxes into correct region proposals for the following YOLO net. Secondly, we design suitable YOLO net for fine detection in the region proposals generated by ACF-PR. Lastly, we design a post-processing step, in which the results of YOLO net are mapped into the original image outputting the detection and localization results. Experiments performed on the Tsinghua-Daimler Cyclist Benchmark with high resolution images and complex scenes show that the proposed method outperforms the other tested representative detection methods in average precision, and that it outperforms YOLOv3 by average precision and outperforms SSD by average precision.