Object Recognition Performance Research Articles

Perinatal Supplementation of Egg Yolk Enhances Cognition and Alters Brain Network Function in Offspring

ObjectivesEgg yolks are a nutrient dense food often recommended to pregnant women, as they contain bioactive compounds that enhance brain development in offspring. We examined the effects of perinatal intake of egg yolks on cognition and brain functional network activity in offspring using a translational sow/piglet dyad model. MethodsSows were fed a control diet (n = 6) or a diet containing egg yolks (n = 5,350 mg egg yolk powder/kg BW/day, equivalent to ∼3 eggs/day for humans) from late gestation throughout lactation. At weaning, piglet offspring (n = 4/sow, total n = 44) performed object recognition testing (ORT) to assess hippocampal-dependent learning and memory outcomes. A subset of the piglets (n = 2/sow, total n = 22) underwent magnetic resonance imaging (MRI) to acquire diffusion tensor imaging (DTI) and resting-state functional MRI. Correlation analysis of outcomes from ORT and MRI was performed. ResultsPiglets from egg yolk fed sows spent more time with novel objects than familiar objects (p = 0.039), suggesting that maternal intake of egg yolk may improve hippocampal-dependent cognitive outcomes. DTI analysis showed that perinatal intake of egg yolks tended to increase fiber length in the hippocampus (p = 0.073) compared to control. Moreover, intake of egg yolk resulted in a 14.3% increase in functional connectivity in the cerebellar network (p = 0.025) and a 15.6% trending increase in the executive network (p = 0.089) compared to control. Interestingly, the visual network was positively correlated with proportional frequency (p = 0.043) and time (p = 0.078) spent with novel objects and negatively correlated with proportional frequency (p = 0.007) and time (p = 0.016) spent with familiar objects, suggesting that the activation of visual network may influence hippocampal-dependent cognitive outcomes at weaning. ConclusionsPerinatal intake of egg yolk enhanced hippocampal-dependent learning and memory, increased fiber length in the hippocampus and altered brain functional connectivity in offspring at weaning. The findings from this study support that egg yolk altered activation of specific brain networks are that associated with cognitive outcomes in weaning piglets. Funding SourcesThis work was supported by grants to H.J.P from the USDA-National Institute of Food and Agriculture and the Egg Nutrition Center.

Micro-Knowledge Embedding for Zero-shot Classification

Zero-shot learning is one of the most challenging machine learning tasks, in which learning stable and transferable knowledge from seen classes plays a pivotal role. To improve the currently unsatisfactory performance of zero-shot object recognition, this paper proposes a novel image representation method, namely, micro-knowledge. In our method, the segmentation of micro-regions and the consequent learning of micro-knowledge are unified by the introduction of a self-attention mechanism. A zero-shot classification framework is carefully designed based on micro-knowledge of images. Under this framework, multiple micro-region descriptions are first obtained by embedding micro-knowledge and then merged to carry out the final classification of unseen objects. Finally, a capsule-unified framework is employed as a graphical programming tool to accomplish the aforementioned tasks. Experiments on public datasets show that the proposed framework can generally achieve competitive results for the classification of unseen objects. Specifically, these results verify that the micro-knowledge learned from one dataset can be directly applied to others without complicated adjustments and demonstrate that using visual features instead of semantic features can result in a decrease in classification error. This research will bring new ideas into the field of zero-shot learning and will serve as an appealing option when addressing the problem of domain shift.

Hovering control of UUV through underwater object detection based on deep learning

It is difficult to find a target object in water and hover an UUV at a relative position from the object to work on it. To solve this problem in this study, we used the geometric principles of camera image mapping to find the real-time position of an underwater object, based on which we investigated the hovering control of an UUV. For underwater object recognition, we used YOLOv2, a deep learning-based object detection algorithm, which has excellent real-time performance. To recognize objects in various underwater environments, the training was conducted on data obtained under different underwater environment conditions, such as illuminance, distance, and obstacles, and the performance of underwater object recognition was increased by using UUV to acquire training data in this study. An UUV was designed and fabricated to implement the proposed algorithm, and a hovering control algorithm was developed for the UUV to recognize a star-shaped object and control its relative distance and direction. The learning data were built, and the object recognition rate was increased using many real sea tests. Moreover, the camera mapping algorithm and the UUV control algorithm proposed in this paper were applied together and tested at sea, to achieve a stable control of the UUV.

Therapeutic Effects of High-Intensity Interval Training Exercise Alone and Its Combination with Ecdysterone Against Amyloid Beta-Induced Rat Model of Alzheimer's Disease: A Behavioral, Biochemical, and Histological Study.

Hippocampal oxidative stress has a vital role in the pathophysiology of Alzheimer's disease (AD)-associated behavioral deficits. Ecdysterone (Ecdy), a natural product and primary steroid hormone, exhibits anti-oxidative and neuroprotective effects. High-intensity interval training (HIIT) has emerged as an effective method for improving physiological brain functions. The present study was designed to investigate the comparative effects of separate and combined HIIT and Ecdy treatment on behavioral functions, hippocampal oxidative status, histological changes in an amyloid-beta (Aβ)-induced rat model of AD. Adult male rats were treated simultaneously with HIIT exercise and Ecdy (10mg/kg/day; P.O.), starting ten days after Aβ-injection, and they continued for eight consecutive weeks. At the end of the treatment course, the behavioral functions of the rats were assessed by commonly-used behavioral paradigms. Subsequently, brain samples were collected for histological analysis and hippocampus samples were collected for biochemical analysis. Results illustrated that Aβ injection impaired learning and memory performances in both novel object recognition and Barnes maze tests, reduced exploratory/locomotor activities in open field test, enhanced anxiety-like behavior in elevated plus-maze (P < 0.05). These behavioral deficits accompanied hippocampal oxidative stress (decreased total antioxidant capacity content and glutathione peroxidase enzyme activity, increased total oxidant status and malondialdehyde level) and neuronal loss in the cerebral cortex and hippocampus in H&E staining (P < 0.05). HIIT and Ecdy improved anxiety-like behavior, attenuated total oxidant status and malondialdehyde, and prevented the neuronal loss (P < 0.05). However, their combination resulted in a more complete and powerful improvement in all the above-mentioned Aβ-related deficits (P < 0.05). Overall, these data provide evidence that a combination of HIIT and Ecdy treatment improves Aβ-induced behavioral deficits, possibly through ameliorating hippocampal oxidative status and preventing neuronal loss.

Escalated (Dependent) Oxycodone Self-Administration Is Associated with Cognitive Impairment and Transcriptional Evidence of Neurodegeneration in Human Immunodeficiency Virus (HIV) Transgenic Rats.

Substance use disorder is associated with accelerated disease progression in people with human immunodeficiency virus (HIV; PWH). Problem opioid use, including high-dose opioid therapy, prescription drug misuse, and opioid abuse, is high and increasing in the PWH population. Oxycodone is a broadly prescribed opioid in both the general population and PWH. Here, we allowed HIV transgenic (Tg) rats and wildtype (WT) littermates to intravenously self-administer oxycodone under short-access (ShA) conditions, which led to moderate, stable, “recreational”-like levels of drug intake, or under long-access (LgA) conditions, which led to escalated (dependent) drug intake. HIV Tg rats with histories of oxycodone self-administration under LgA conditions exhibited significant impairment in memory performance in the novel object recognition (NOR) paradigm. RNA-sequencing expression profiling of the medial prefrontal cortex (mPFC) in HIV Tg rats that self-administered oxycodone under ShA conditions exhibited greater transcriptional evidence of inflammation than WT rats that self-administered oxycodone under the same conditions. HIV Tg rats that self-administered oxycodone under LgA conditions exhibited transcriptional evidence of an increase in neuronal injury and neurodegeneration compared with WT rats under the same conditions. Gene expression analysis indicated that glucocorticoid-dependent adaptations contributed to the gene expression effects of oxycodone self-administration. Overall, the present results indicate that a history of opioid intake promotes neuroinflammation and glucocorticoid dysregulation, and excessive opioid intake is associated with neurotoxicity and cognitive impairment in HIV Tg rats.

Electrical Capacitance Tomography Sensor With House Structure for Assisting Recognition of Objects

As a fast and non-invasive visualization technology, electrical capacitance tomography(ECT) avoids the problem that camera imaging technology is interfered by light and can’t identify material properties. However, the traditional ECT sensor research mainly focused on the image reconstruction of multiphase flow between electrodes, without considering the object recognition performance through the dielectric distribution of the objects. Therefore, we propose a new ECT sensor (called House Sensor) with house-like structure, which is suitable for multi features detection of household objects. We analyze the sensitivity distribution of House Sensor, and use linear back projection (LBP) and Tikhonov regularization for high-speed image reconstruction. 20 objects with different sizes and dielectric constants are visually clustered by t-distributed stochastic neighbor embedding (t-SNE) to verify the feasibility of classification, and the accuracy of Random Forest (RF) is up to 95.3% within 3.5ms. This implies that House Sensor provide a new feasible scheme to assist object recognition under non-vision conditions.

Sex differences in behavioral pathology induced by subconvulsive stimulation during early postnatal life are overcome by epileptic activity in the pre-juvenile weanling period

Chronic subconvulsive activity in early life leads to sex-related autistic-like deficits in handling, object recognition, and social performance in pre-pubertal rats. Since autism and epilepsy are common neurodevelopmental disorders with high coincidence, we tested whether early-life chronic subconvulsive activity compared to convulsive activity alters handling, spatial memory, lateralization, coping strategy and the seizure threshold in a sex-dependent manner. A hypothesis is that convulsive seizures may alter sex differences induced by subconvulsive (SC) activity. Serial subconvulsive doses of kainic acid (KA) were administered postnatally (0.25–1 mg/kg) for 15 days to induce the chronic subconvulsive phenotype (SC group). Age-matched controls and a subset of SC pups were exposed to a convulsive dose of KA (KA and SC + KA groups; 7.5 mg/kg) or flurothyl vapors. In our open handling test, controls and the ASD groups escaped to a similar degree whereas after convulsive seizures, the pups exhibited freezing behavior; no escapes occurred. In the spontaneous alternating T-Maze control males and females entered the left arm with higher frequency. The SC males but not SC females entered left and right arms to a similar degree; alternation rates were reduced to chance revealing a sex difference. However, in KA and SC + KA groups, there was a sharp loss of spontaneous alternation rates. The rapid repetitive entries shifted to the right in both sexes possibly be due to hippocampal injury and changes in network activity induced by status epilepticus. In the forced swim test (FST), control and CS females were more active than corresponding males. After convulsions, immobility was reduced and vertical mobility was increased in SC and SC + KA males suggesting an elevated coping strategy compared to females. Onset and severity of KA induced status epilepticus was delayed in SC males and females possibly due to desensitization of KA receptors. Following flurothyl exposure, control males had faster onset of twitches and clonic seizures than control females which disappeared after the sub-convulsive pre-treatment. Data suggest that behavioral manifestations are more readily detectable between males and females when low levels of hyperexcitation are present chronically in early postnatal development but diminished after tonic-clonic convulsions persist. Therefore, therapeutic interventions may benefit patients if initiated upon the initial onset of sex-related autistic pathologies, particularly in males, which may reduce subsequent vulnerability to seizures.

The role of ventral stream areas for viewpoint-invariant object recognition

Object constancy is one of the most crucial mechanisms of the human visual system enabling viewpoint invariant object recognition. However, the neuronal foundations of object constancy are widely unknown. Research has shown that the ventral visual stream is involved in processing of various kinds of object stimuli and that several regions along the ventral stream are possibly sensitive to the orientation of an object in space. To systematically address the question of viewpoint sensitive object perception, we conducted a study with stroke patients as well as an fMRI experiment with healthy participants applying object stimuli in several spatial orientations, for example in typical and atypical viewing conditions. In the fMRI experiment, we found stronger BOLD signals and above-chance classification accuracies for objects presented in atypical viewing conditions in fusiform face sensitive and lateral occipito-temporal object preferring areas. In the behavioral patient study, we observed that lesions of the right fusiform gyrus were associated with lower performance in object recognition for atypical views. The complementary results from both experiments emphasize the contributions of fusiform and lateral-occipital areas to visual object constancy and indicate that visual object constancy is particularly enabled through increased neuronal activity and specific activation patterns for objects in demanding viewing conditions.

Shear-Force Sensor With Point-Symmetric Electrodes Driven by LTPS TFT Active Matrix Backplane

In this study, we successfully demonstrated a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$700 ~\mu \text{m}$ </tex-math></inline-formula> thick shear-force sensor with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$60\times60$ </tex-math></inline-formula> high resolution array by introducing a combination of an originally developed pressure-sensing elastomer and a newly proposed taxel (tactile pixel) structure with four point-symmetric electrodes. When shear forces are applied to the sensor substrate in several directions by lifting a weight, four sets of resistances between the electrodes are read through low-temperature polycrystalline-silicon (LTPS) thin-film transistors (TFTs) in each taxel and clear outputs corresponding to that direction can be confirmed. Furthermore, as a result of evaluating the object recognition performance of the tactile sensor by using jigs with engraved small dimples, it was confirmed that the recognized minimum size of the shape was 1.5 mm, which suggests that the spatial resolution of this sensor is superior to the human hand perception ability. Because this sensor is ultra-thin, not bulky, and potentially applicable to flexible substrates, it is quite promising as a sensor that can be mounted on the fingertip of a robot hand with multiple functions.

Attentional Features of Mindfulness are Better Predictors of Face Recognition than Empathy and Compassion-Based Constructs.

Recent research has employed measures of either empathy, compassion or mindfulness and linked better face recognition memory to higher scores of identification with all humanity and mindfulness but not empathy or compassion. Additionally, empathy, compassion and mindfulness have been suggested as concepts that intertwine, but research has not yet examined how their respective personality questionnaires map onto latent concepts. We employed these measures together to explore their factor structure and, using structural equation modelling, we investigated if the suggested latent variables predict recognition memory performance for face and non-face stimuli. Attentional notions of mindfulness described a latent factor that predicted face recognition. All self-compassion facets and the non-react mindfulness facet described a latent factor, which predicted false alarms in face recognition. Finally, empathy and compassion-based notions described one latent factor, which did not predict recognition performance. None of the latent variables predicted performance in either object or voice recognition. Collectively, findings indicate attention-based mindfulness to benefit face recognition, prompting further research into the potential of mindfulness to support the face recognition process.

PDE: A Real-Time Object Detection and Enhancing Model under Low Visibility Conditions

Deep object detection models are important tools that can accurately detect objects and frame them for the user in real time. However, in low visibility conditions, such as fog or low light conditions, the captured images are underexposed and blurred, which negatively affects the recognition accuracy and is not well visible to humans. In addition, the image enhancement model is complex and time-consuming. Using the image enhancement model before the object recognition model cannot meet the real-time requirements. Therefore, we propose the Parallel Detection and Enhancement model (PDE), which detects objects and enhances poorly visible images in parallel and in real time. Specifically, we introduce the specially designed tiny prediction head along with coordinated attention and multi-stage concatenation modules to better detect underexposed and blurred objects. For the parallel image enhancement model, we adaptively develop improved weighting evaluation models for each “3D Lookup Table” module. As a result, PDE achieves better detection accuracy for poorly visible objects and more user-friendly reference in real time. Experimental results show that PDE has significantly better object recognition performance than the state-of-the-art on real foggy (8.9%) and low-light (20.6%) datasets.

Distinguishing transient from persistent tactile agnosia after partial anterior circulation infarcts – Behavioral and neuroimaging evidence for white matter disconnection

From a cohort of 36 patients presenting apperceptive tactile agnosia after first cortical ischemic stroke, 14 showed temporary impairment at admission. A previous multi-voxel analysis of the cortical lesions, using as explanatory variable the course of tactile object recognition performance over the recovery period of 9 months, partitioned the cohort into three subgroups. Of the 14 patients constituting two of the subgroups, 7 recovered from their impairment whereas 7 did not. These two subgroups could not be distinguished at admission. The primary aim of the present study is to present two assessments that can do so.The first assessment comprises a pattern of behavioral measures, determined via principal component analysis, encoded in three tests: picking small objects, macrogeometrical discrimination and tactile object recognition. The receiver operating characteristic curve derived from permutation of the behavioral test scores yielded an 80% probability of correct identification of the patient subgroup and an 8% probability for false identification. As done with the permuted scores, the pattern could predict the persistence of affliction of new stroke patients with tactile agnosia.The second predictive assessment extends our previous evaluation of cortical MRI lesion maps to include subcortical regions. Confirming our previous study, the lesions of the persistently impaired subgroup disrupted significantly the anterior arcuatus fasciculus and associated superior longitudinal fasciculus III in the ipsilesional hemisphere, impeding reciprocal information transfer between supramarginal gyrus and both the ventral premotor cortex and Brodmann area 44. Due to the importance of interhemispheric information transfer in tactile agnosia, we performed a supplementary analysis of tactile object recognition scores. It showed that haptic information transfer from the non-affected to the affected hands in the persistent cases partly restored function during the nine months, possibly following restoration of functional interhemispheric haptic information transfer at the border of posterior corpus callosum and splenium.In conclusion, the combined findings of the cortical lesion at subarea PFt of the inferior parietal lobule and the associated subcortical tract lesions permit almost perfect prediction of persistent impairment of tactile object recognition. The study substantiates the need for combined analysis of both cortical lesions and white matter tract disconnections.

Teaching Teachers First and Then Student: Hierarchical Distillation to Improve Long-Tailed Object Recognition in Aerial Images

Remote sensing data distribution generally exposes the long-tail characteristic. This will limit the object recognition performance of existing deep models when they are trained with such unbalanced data. In this paper, we propose a novel hierarchical distillation framework to address the long-tailed object recognition in aerial images. Firstly, we notice that not only student model should learn feature representations from teachers, but also teacher models should learn feature representations from each other. Therefore, we build hierarchical teacher-wise distillation to improve the feature representations of the teacher models trained with middle and tail data, which is achieved by distilling the feature representations of the teacher model trained with head data. Secondly, we notice that the feature representations of the middle and tail classes can not be effectively distilled from the teacher to the student, since too little middle and tail data can be used to learn. Thus, we propose self-calibrated sampling learning that enforces the student to strengthen the learning of the middle and tail data, thereby improving the student&#x2019; feature learning ability. Extensive experiments on two widely-used DOTA and FGSC-23 datasets demonstrate superior performance of the proposed method compared with state-of-the-art methods. Model and code are publicly available at: https://github.com/wdzhao123/T2FTS.

Real-Time Power Line Detection for Safe Flight of Agricultural Spraying Drones Using Embedded Systems and Deep Learning

Power line detection is necessary for the safe flight of low-flying UAVs (Unmanned Aerial Vehicles). This paper deals with the power line recognition problem for the safety of agricultural spraying drones in agricultural environments. The dataset of power lines was obtained in an agricultural environment. The training dataset was constructed by labeling powerlines with bounding boxes of 6 sizes, ranging from 0.03 to 0.15 times the image. The model used for training was the tiny-YOLOv3 model. The model was verified using the mean average precision (mAP), which was used to verify the object recognition performance. Depending on the six sizes of bounding boxes, the mAPs were evaluated to be 70.22, 94.00, 86.75, 68.87, 61.65, and 53.40, respectively. The mAP was the highest at the bounding box of 0.05 times the image size, and it was confirmed that this size is most suitable for power line detection. The real-time frames per second (FPS) results of power lines detection are on average 12.5. This paper shows that the location detection of power lines is possible in real-time using deep-learning techniques with embedded systems.

PROTOTIPE APLIKASI PENGENALAN WAYANG KULIT MENGGUNAKAN CNN BERBASIS VGG16

Indonesia has various types of culture and traditional arts. In this era of globalization, local culture and arts have begun to be eroded by the times. One of the diverse Indonesian culture is wayang kulit. Where the shadow puppets in Indonesia vary and vary from region to region. In this case, the puppet characters have different forms and curves, so recognizing the shape of a puppet is very difficult. In the development of technology, computer vision technology began to be widely used to perform object recognition with deep learning learning. So that an object being studied can be detected properly. In this study, a prototype was made with the detection of puppet types using Deep Learning learning using Convolutional Neural Networks to detect shadow puppet objects based on the VGG16 architecture. The results obtained by the CNN and VGG16 methods reached 86%. With the results obtained, a prototype model is made which will later be able to help the community in the introduction of shadow puppets.Keyword: CNN, shadow puppets ,VGG16

Differential association of baseline body weight and body-weight loss with neurological deficits, histology, and death after repetitive closed head traumatic brain injury

Clinical observations indicate that body weight (BW) extremes are associated with worse outcome after traumatic brain injury (TBI); yet, it is uncertain whether the baseline BW (bBW) may affect outcome after mouse TBI. We retrospectively analyzed 129 similarly aged (9–12 weeks) male C57BL6/J mice that were subjected to repetitive closed head TBI (rTBI) using an established weight drop paradigm as well as 55 sham injured mice. We sought to determine whether the bBW as well as the post-TBI weight relative to baseline (%BW) were associated with a variety of post-rTBI outcomes, including acute model complications (skull fractures and macroscopic hemorrhage), impact seizures, return of the righting reflex (RR), the neurological severity score (NSS), post-rTBI BW-change, and 28-day mortality. In a subset of rTBI mice, we also assessed for potential associations between the bBW and %BW and performance in the novel object recognition (NOR) task and various histological outcomes at 28 days. We found no association between the bBW with acute model complications, impact seizure burden, RR, NSS, and NOR performance at 28 days, as well as cerebral microbleed burden, presence of hyperphosphorylated tau, and TDP-43 pathology after rTBI. However, a higher bBW was associated with a longer time to first impact seizure, a greater microglial activation, astrocytosis, and neuronal loss in the injured cerebral cortex at 28 days. A greater %BW-loss was associated with a shorter impact seizure-free survival, longer time to return of the righting reflex, greater neurological deficit severity as assessed by the NSS and NOR, and worse mortality. On multiple linear regression there was no independent association of the %BW-loss with neuronal loss and neuroinflammation after adjustment for the bBW. These observations indicate that the bBW and %BW-loss may be important biological variables in certain experimental mouse TBI investigations, depending on the outcome measures of interest.

Neurobehavioral effects of chronic low-dose vanadium administration in young male rats

Exposure to the metal vanadium, in both animals and humans has been linked to various physiological consequences including respiratory and gastrointestinal conditions. Research on the neurobehavioral effects of vanadium exposure is limited. Hence, the purpose of the current study was to examine the effects of chronic low-dose vanadium administration (0.04 mg/week) on the behavior of young male rats. Four weeks following the administration of vanadium, rats were tested on the open field, object recognition, and Morris Water maze tasks. Vanadium did not affect exploration, locomotion, or anxiety-like behavior as measured by the open field task. Vanadium administration affected novel object recognition performance. Intriguingly, rats exposed to vanadium exhibited lower latency times on day 2 of the Morris Water maze. These findings suggest that vanadium’s behavioral effects are complex and warrant further investigation to better understand the potential benefits and consequences of its exposure.

OrthographicNet: A Deep Transfer Learning Approach for 3-D Object Recognition in Open-Ended Domains

Nowadays, service robots are appearing more and more in our daily life. For this type of robot, open-ended object category learning and recognition is necessary since no matter how extensive the training data used for batch learning, the robot might be faced with a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">new object</i> when operating in a real-world environment. In this article, we present <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OrthographicNet</i> , a convolutional neural network based model, for 3-D object recognition in open-ended domains. In particular, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OrthographicNet</i> generates a global rotation- and scale-invariant representation for a given 3-D object, enabling robots to recognize the same or similar objects seen from different perspectives. Experimental results show that our approach yields significant improvements over the previous state-of-the-art approaches concerning object recognition performance and scalability in open-ended scenarios. Moreover, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OrthographicNet</i> demonstrates the capability of learning new categories from very few examples on-site. Regarding real-time performance, three real-world demonstrations validate the promising performance of the proposed architecture.

Automated interpretation and analysis of bronchoalveolar lavage fluid

BackgroundThe cytological analysis of bronchoalveolar lavage fluid (BALF) plays an essential role in the differential diagnosis of respiratory diseases. In recent years, deep learning has demonstrated excellent performance in image processing and object recognition. ObjectivesWe aim to apply deep learning to the automated interpretation and analysis of BALF. MethodVisual images were acquired using an automated biological microscopy platform. We propose a three-step algorithm to automatically interpret BALF cytology based on a convolutional neural network (CNN). The clinical value was evaluated at the patient level. ResultsOur model successfully detected most cells in BALF specimens and achieved a sensitivity, precision, and F1 score of over 0.9 for most cell types. In two tests in the clinical context, the algorithm outperformed experienced practitioners. ConclusionThe program can automatically provide the cytological background of BALF and augment clinical decision-making for clinicians.

Incoherent reconstruction-free object recognition with mask-based lensless optics and the Transformer.

A mask-based lensless camera adopts a thin mask to optically encode the scene and records the encoded pattern on an image sensor. The lensless camera can be thinner, lighter and cheaper than the lensed camera. But additional computation is required to reconstruct an image from the encoded pattern. Considering that the significant application of the lensless camera could be inference, we propose to perform object recognition directly on the encoded pattern. Avoiding image reconstruction not only saves computational resources but also averts errors and artifacts in reconstruction. We theoretically analyze multiplexing property in mask-based lensless optics which maps local information in the scene to overlapping global information in the encoded pattern. To better extract global features, we propose a simplified Transformer-based architecture. This is the first time to study Transformer-based architecture for encoded pattern recognition in mask-based lensless optics. In the optical experiment, the proposed system achieves 91.47% accuracy on the Fashion MNIST and 96.64% ROC AUC on the cats-vs-dogs dataset. The feasibility of physical object recognition is also evaluated.