The object detection task is usually affected by complex backgrounds. In this paper, a new image object detection method is proposed, which can perform multi-feature selection on multi-scale feature maps. By this method, a bidirectional multi-scale feature fusion network is designed to fuse semantic features and shallow features to improve the detection effect of small objects in complex backgrounds. When the shallow features are transferred to the top layer, a bottom-up path is added to reduce the number of network layers experienced by the feature fusion network, reducing the loss of shallow features. In addition, a multi-feature selection module based on the attention mechanism is used to minimize the interference of useless information on subsequent classification and regression, allowing the network to adaptively focus on appropriate information for classification or regression to improve detection accuracy. Because the traditional five-parameter regression method has severe boundary problems when predicting objects with large aspect ratios, the proposed network treats angle prediction as a classification task. The experimental results on the DOTA dataset, the self-made DOTA-GF dataset and the HRSC 2016 dataset show that, compared with several popular object detection algorithms, the proposed method has certain advantages in detection accuracy.

Protrusive defects on the color filter of thin-film transistor (TFT) liquid crystal displays (LCDs) frequently damage the valuable photomask. An fast method using side-view illuminations associated with digital charge-couple devices (CCDs) to detect the protrusive defect in the four substrates, which are the black matrix (BM), red, green, and blue. Between the photomask and substrate, the depth of field (DOF) is normally 300 μm for the proximity-type aligner; we select the four substrates to evaluate the detectability in the task. The experiment is capable of detecting measurements of 300 μm and even lower than 100 μm can be assessed successfully. The maximum error of the measurement is within 6% among the four samples. Furthermore, the uncertainty analysis of three standard deviations is conducted. Thus, the method is cost-effective to prevent damage for valuable photomasks in the flat panel display industry.

This paper proposes an encryption scheme for high pixel density images. Based on the application of the quantum random walk algorithm, the Long short-term memory (LSTM) can effectively solve the problem of low efficiency of the quantum random walk algorithm in generating large-scale pseudorandom matrices, and further improve the statistical properties of the pseudorandom matrices required for encryption. The LSTM is then divided into columns and fed into the LSTM in order for training. Due to the randomness of the input matrix, the LSTM cannot be trained effectively, so the output matrix is predicted to be highly random. The LSTM prediction matrix of the same size as the key matrix is generated based on the pixel density of the image to be encrypted, which can effectively complete the encryption of the image. In the statistical performance test, the proposed encryption scheme achieves an average information entropy of 7.9992, an average number of pixels changed rate (NPCR) of 99.6231%, an average uniform average change intensity (UACI) of 33.6029% and an average correlation of 0.0032. Finally, various noise simulation tests are also conducted to verify its robustness in real-world applications where common noise and attack interference are encountered.

The quantification of sea surface temperature (SST) through space platforms has revolutionized how we obtain information at a global level. However, the main disadvantage of obtaining SST with satellite images consists of its inherent coarse spatial resolution. One solution could be the use of downscaling algorithms to create sequences of matrices at a higher resolution. We used the same SST data source from the MODIS-Aqua sensor at three spatial resolutions of 9 km, 4.5 km, and 1 km in the Gulf of California. Based on an open-source algorithm, the original SST images were downscaled to 4.5 km, 1 km, 500 m, 250 m, and 125 m per pixel scales. Results indicate a strong linear relationship between the original SST-MODIS data and the modeled data for all spatial resolutions. This study demonstrates the feasibility of an open-source downscaling algorithm to enhance the spatial resolution of SST images in a marginal sea.

Obtaining classification confidence at the pixel level is a challenging task for accuracy assessment in remote sensing image classification. Among the various methods for estimating classification confidence at the pixel level, interpolation-based methods have drawn special attention in the literature. Even though they have been widely recognized in the literature, their usefulness has not been rigorously evaluated. This paper conducts a comprehensive evaluation of three interpolation-based methods: local error matrix method, bootstrap method, and geostatistical method. We applied each of the three methods to three representative datasets with different spatial resolutions, spectral bands, and the number of classes. We then derive the estimated classification confidence and true classification confidence and compared the results with each other using both exploratory data analysis (bi-histogram) and statistical analysis (Willmott's d and Binned classification quality). The results indicate that the three interpolation methods provide some interesting insights on various aspects of estimating per-pixel classification confidence. Unfortunately, the interpolation assumes that classification confidence is smooth across the space, which is usually not true in practice. In other words, interpolation-based methods have limited practical use.

In this contribution, we present a high-speed multiplex grating spectrometer based on a spectral coding approach that is founded on principles of compressive sensing. The spectrometer employs a single-pixel InGaAs detector to measure the signals encoded by an amplitude spatial light modulator (digital micromirror device, DMD). This approach leads to a speed advantage and multiplex sensitivity advantage atypical for standard dispersive systems. Exploiting the 18.2 kHz pattern rate of the DMD, we demonstrate 4.2 ms acquisition times for full spectra with a bandwidth of 450 nm (5250 cm-1 – 4300 cm-1; 1.9 µm – 2.33 µm). Due to the programmability of the DMD, spectral regions of interest can be chosen freely, thus reducing acquisition times further, down to the sub-millisecond regime. The adjustable resolving power of the system accessed by means of computer simulations is discussed, quantified for different measurement modes, and verified by comparison with a state-of-the-art Fourier-transform infrared spectrometer. We show measurements of characteristic polymer absorption bands in different operation regimes of the spectrometer. The theoretical multiplex advantage of 8 was experimentally verified by comparison of the noise behavior of the spectral coding approach and a standard line-scan approach.

Effectively mapping and monitoring rubber plantation is still changing. Previous studies have explored the potential of phenology features for rubber plantation mapping through a pixel-based approach (pixel-based phenology approach). However, in fragmented mountainous Xishuangbanna, it could lead to noises and low accuracy of resultant maps. In this study, we investigated the capability of an integrated approach by combining phenology information with an object-based approach (object-based phenology approach) to map rubber plantations in Xishuangbanna. Moderate Resolution Imaging Spectroradiometer (MODIS) data were firstly used to acquire the temporal profile and phenological features of rubber plantations and natural forests, which delineates the time windows of defoliation and foliation phases. Landsat images were then used to extract a phenology algorithm comparing three different approaches: pixel-based phenology, object-based phenology, and extended object-based phenology to separate rubber plantations and natural forests. The results showed that the two object-based approaches achieved higher accuracy than the pixel-based approach, having overall accuracies of 96.4%, 97.4%, and 95.5%, respectively. This study proved the reliability of a phenology-based rubber mapping in fragmented landscapes with a distinct dry/cool season using Landsat images. This study indicated that the object-based phenology approaches can effectively improve the accuracy of the resultant maps in fragmented landscapes.

Wafer-scale blue micro-light-emitting diode (micro-LED) arrays are fabricated with a pixel size of 12 μm, a pixel pitch of 15 μm, and a pixel density of 1692 pixels per inch, achieved by optimizing the properties of e-beam-deposited and sputter-deposited indium tin oxide (ITO). Although the sputter-deposited ITO (S-ITO) films exhibit a densely packed morphology and lower resistivity compared to those of the e-beam-deposited ITO (E-ITO) films, the forward voltage (VF) values of a micro-LED with the S-ITO films are higher than those with the E-ITO films. The VF values for a single pixel and for four pixels with E-ITO films are 2.82 V and 2.83 V, respectively, while the corresponding values for S-ITO films are 3.50 V and 3.52 V. This was attributed to ion bombardment damage and nitrogen vacancies in the p-GaN layer. Surprisingly, the VF variations of a single pixel and of four pixels with the optimized E-ITO spreading layer from five different regions are only 0.09 V and 0.10 V, respectively. This extremely uniform VF variation is suitable for realizing micro-LED displays to be used in AR and VR applications, circumventing the bottleneck in the development of long-lifespan and high-brightness organic LED devices for industrial mass production.

There is a need for cost-effective and non-invasive methods of monitoring feeding behavior in livestock operations, considering the significant impact of feed costs on economic efficiency and assisting in detecting health issues of group-fed animals. This paper proposes using deep learning-based computer vision techniques to detect pen-fed beef cattle feeding behavior using Mask Region-based Convolutional Neural Network (RCNN). A deep learning model was pre-trained on the Common Objects in Context (COCO) dataset to generate cattle instance segmentation. Manually defined feed bunk polygons are compared with these segmentation masks to derive feeding time for each bunk. A full day’s worth of video data and the corresponding physical sensor data are collected for the experiment. By benchmarking the computer vision detected data with physical ground truth over random time segments from morning to evening (thus various lighting conditions), the optimal thresholds for Mask RCNN are determined to be 0.7 for bounding boxes and 0.1 for masks. Using these parameters. The reports suggest that the computer vision system achieved a precision of 87.2% and a recall of 89.1%, signifying precise detection of feeding events. Our study, to the best of our knowledge, was one of the first investigations of instance segmentation on feeding time sense, which combines deep learning methods with traditional computer vision logistics, reporting on feeding time data collection and processing, camera testing and adjustment, and performance evaluation. Future research directions include computer vision applied in feed grading and animal re-identification for individual production analysis.

Natural image matting is an essential technique for image processing that enables various applications, such as image synthesis, video editing, and target tracking. However, the existing image matting methods may fail to produce satisfactory results when computing resources are limited. Sampling-based methods can reduce the dimensionality of the decision space and therefore reduce computational resources by employing different sampling strategies. While these approaches reduce computational consumption, they may miss an optimal pixel pair when the number of available high-quality pixel pairs is limited. To address this shortcoming, we propose a novel multi-criterion sampling strategy that avoids missing high-quality pixel pairs by incorporating multi-range pixel pair sampling and high-quality samples selection method. This strategy is employed to develop a multi-criterion matting algorithm via Gaussian process, which searches for the optimal pixel pair by using the Gaussian process fitting model instead of solving the original pixel pair objective function. Experimental results demonstrate that our proposed algorithm outperforms other methods even with 1% computing resources, and achieves alpha matte results comparable to those yielded by the state-of-the-art optimization algorithms.

The processing of cloud free geo-referenced imagery is one of the preliminary processing step of any land application. This letter describe the methodology developed to obtain a seamless cloud free composite of Africa for 2016 using Sentinel-2A data at 10 meters resolution freely available from the European Space Agency. The method is based on an hybrid method resulting from the merging of the two most robust time series methods namely the "darkest pixel" and the "maximum NDVI" previously developed with AVHRR time series.

Segmentation of a region of interest is an important pre-processing step for many colour image analysis techniques. Similarly segmentation of plant in digital images is an important preprocessing step in phenotying plants by image analysis. In this paper we present an analytical study to statistically determine the suitability of colour space representation of an image to best detect plant pixels and separate them from background pixels. Our hypothesis is that the colour space representation in which the separation of the distributions representing plant pixels and background pixels is maximized would be the best for detection of plant pixels. The two classes of pixels are modelled as a Gaussian mixture model (GMM). In our GM modelling we don't make any prior assumption about the number of Gaussians in the model. Rather a constant bandwidth mean-shift filter is used to cluster the data and the number of clusters and hence the number of Gaussians is automatically determined. Here we have analysed following representative colour spaces like $RGB$, $rgb$, $HSV$, $Ycbcr$ and $CIE-Lab$. This is because these colour spaces represent several other similar colour spaces and also an exhaustive study of all the colour space will be too voluminous. We also analyse the colour space feature from the two-class variance ratio perspective and compare the results of our hypothesis with this metric. The dataset for this empirical study consist of 378 digital images of plants and their manual segmentation. Dataset consist of various species of plants (arabidopsi, tobacco, wheat, rye grass etc.) imaged under different lighting conditions, indoor and outdoor, controlled and uncontrolled background. In results we obtain better segmentation of the plants in $HSV$ colour space, which is supported by its Earth mover distance (EMD) on the GMM distribution of plant and background pixels.

In space application, hybrid pixel detectors of the Timepix family have been considered mainly for measurement of radiation levels and radiation dosimetry in low earth orbits. By the example of the Space Application of Timepix Radiation Monitor (SATRAM), we demonstrate the unique capabilities of Timepix-based miniaturized radiation detectors for particle separation. Using a novel method for proton spectrum reconstruction, we were able to measure the spectrum of protons trapped in the inner Van-Allen radiation belt, for the first time with a single-layer detector. We assess the measurement stability and the resiliency of the detector to the space environment demonstrating that, even though degradation is observed, data quality has not been affected significantly over more than 10 years. Based on the SATRAM heritage and the capabilities of the latest-generation Timepix-series chips, we discuss their applicability for use in a compact magnetic spectrometer for deep-space mission or in the high radiation environment of the Jupiter radiation belts, and their capability for use as single-layer X- and γ-ray polarimeter. The latter was supported by measurement of the polarization of scattered radiation in a laboratory experiment, where a modulation of 80% was found.

The sPHENIX experiment is a state-of-the-art jet and heavy flavor physics detector which successfully recorded its first Au + Au collision data at 200 GeV at RHIC. sPHENIX will provide heavy flavor physics measurements covering an unexplored kinematic region and unprecedented precision at RHIC to probe the parton energy loss mechanism, parton transport coefficients in quark-gluon plasma, and the hadronization process under various medium conditions. At the center of sPHENIX, the Monolithic-Active-Pixel-Sensor (MAPS) based VerTeX detector (MVTX) is a high-precision silicon pixel detector. The MVTX provides excellent position resolution and the capability of operating in continuous streaming readout mode, allowing precise vertex determination and recording a large data sample, both of which are particularly crucial for heavy flavor physics measurements. In this work, we will show the general performance of heavy-flavor hadron reconstruction. In addition, we will discuss the commissioning experience with sPHENIX. Finally, we will provide the projection of b-hadron and jet observables and discuss the estimated constraints on theoretical models.

It is very difficult to accurately divide farmland and woodland in Gaofen 2 (GF-2) remote sensing image, because their single plant coverage is very small, and their spectra are very similar. The ratio of spatial resolution and one plant’s coverage area must be fully taken into account when designing the Convolutional Neural Network structure for extracting them from GF-2 image. We establish a Convolutional Encode Neural Networks model (CENN), The first layer has two sets of convolution kernels to learn the characteristics of farmland and woodland respectively, while the second layer is the encoder to encode the characteristics by transfer function, which can map the results to the corresponding category number. In the training stage, samples of farmland, woodland, and other categories are categorically used to train CENN, as soon as training is accomplished, CENN would acquire enough ability to accurately extract farmland and woodland from remote sensing images. The final extraction result is obtained by implementing per-pixel segmentation of images used to train the CENN. CENN is compared and analyzed with others such as Deep Belief Network (DBN), Full Convolutional Network (FCN), Deeplab Model. The results of experiments show that CENN can more accurately mine the characteristics of farmland and woodland, and it achieves its goal of extracting farmland and woodland with high precision from GF-2 images.

This research aims to estimate the micro-motion (m-m) of ships. The problem of motion and m-m detection of targets is usually solved using synthetic aperture radar (SAR) along-track interferometry (ATI) which is observed employing two radars spatially distanced by a baseline extended in the azimuth direction. This paper is proposing a new approach where the m-m estimation of ships, occupying thousands of pixels, is measured processing the information given by sub-pixel tracking generated during the coregistration process of several re-synthesized time-domain and overlapped sub-apertures. The SAR products are generated splitting the raw data, according to a small-temporal baseline strategy, observed by one single wide-band staring spotlight (ST) SAR image. The predominant vibrational modes of different ships are estimated and results are promising to extend this application in performing surveillance also of land-based industries activities. Experiments are performed processing one ST SAR image observed by the COSMO-SkyMed satellite system.