Visual Judgment Research Articles

Utility of U-Net for the objective segmentation of the fibroglandular tissue region on clinical digital mammograms

This study investigates the equivalence or compatibility between U-Net and visual segmentations of fibroglandular tissue regions by mammography experts for calculating the breast density and mean glandular dose (MGD). A total of 703 mediolateral oblique-view mammograms were used for segmentation. Two region types were set as the ground truth (determined visually): (1) one type included only the region where fibroglandular tissue was identifiable (called the ‘dense region’); (2) the other type included the region where the fibroglandular tissue may have existed in the past, provided that apparent adipose-only parts, such as the retromammary space, are excluded (the ‘diffuse region’). U-Net was trained to segment the fibroglandular tissue region with an adaptive moment estimation optimiser, five-fold cross-validated with 400 training and 100 validation mammograms, and tested with 203 mammograms. The breast density and MGD were calculated using the van Engeland and Dance formulas, respectively, and compared between U-Net and the ground truth with the Dice similarity coefficient and Bland–Altman analysis. Dice similarity coefficients between U-Net and the ground truth were 0.895 and 0.939 for the dense and diffuse regions, respectively. In the Bland–Altman analysis, no proportional or fixed errors were discovered in either the dense or diffuse region for breast density, whereas a slight proportional error was discovered in both regions for the MGD (the slopes of the regression lines were −0.0299 and −0.0443 for the dense and diffuse regions, respectively). Consequently, the U-Net and ground truth were deemed equivalent (interchangeable) for breast density and compatible (interchangeable following four simple arithmetic operations) for MGD. U-Net-based segmentation of the fibroglandular tissue region was satisfactory for both regions, providing reliable segmentation for breast density and MGD calculations. U-Net will be useful in developing a reliable individualised screening-mammography programme, instead of relying on the visual judgement of mammography experts.

Moving towards intelligent telemedicine: Computer vision measurement of human movement

Background:Telemedicine video consultations are rapidly increasing globally, accelerated by the COVID-19 pandemic. This presents opportunities to use computer vision technologies to augment clinician visual judgement because video cameras are so ubiquitous in personal devices and new techniques, such as DeepLabCut (DLC) can precisely measure human movement from smartphone videos. However, the accuracy of DLC to track human movements in videos obtained from laptop cameras, which have a much lower FPS, has never been investigated; this is a critical gap because patients use laptops for most telemedicine consultations. Objectives:To determine the validity and reliability of DLC applied to laptop videos to measure finger tapping, a validated test of human movement. Method:Sixteen adults completed finger-tapping tests at 0.5 Hz, 1 Hz, 2 Hz, 3 Hz and at maximal speed. Hand movements were recorded simultaneously by a laptop camera at 30 frames per second (FPS) and by Optotrak, a 3D motion analysis system at 250 FPS. Eight DLC neural network architectures (ResNet50, ResNet101, ResNet152, MobileNetV1, MobileNetV2, EfficientNetB0, EfficientNetB3, EfficientNetB6) were applied to the laptop video and extracted movement features were compared to the ground truth Optotrak motion tracking. Results:Over 96% (529/552) of DLC measures were within +/−0.5 Hz of the Optotrak measures. At tapping frequencies >4 Hz, there was progressive decline in accuracy, attributed to motion blur associated with the laptop camera’s low FPS. Computer vision methods hold potential for moving us towards intelligent telemedicine by providing human movement analysis during consultations. However, further developments are required to accurately measure the fastest movements.

The Effect of Odor Valence on Facial Attractiveness Judgment: A Preliminary Experiment

The role of social odors on human social interactions, including face evaluation, has been widely indicated. However, for nonsocial odors, there has not been a consistent conclusion. Therefore, this study aimed to verify the effect of suprathreshold nonsocial odors on facial attractiveness judgment when the visual input is ambiguous. We designed a 3 (odor valence: neutral, pleasant, and unpleasant) × 7 (continuous levels of morphed fuzziness of attractiveness: 37.5% to 62.5%) within-subject experiment. A total of 30 participants (18 females) completed the whole experiment simultaneously for three consecutive days. The results showed that faces presented with pleasant and neutral odors were judged as significantly more attractive than those with unpleasant odors. The intervention effect of odor valence on facial attractiveness differed by fuzzy attractiveness levels. Results also suggested that male faces were perceived as more attractive than female faces no matter the odor conditions. The results of this study provide evidence to support the cross-modal emotion integration effect of olfaction and vision. Follow-up studies need to be conducted to reveal the underlying mechanism of odor valence on visual fact attractive judgment.

Development of an Artificial Intelligence Tool for Intraoperative Guidance During Endovascular Aneurysm Repair

Adverse events during surgery can occur in part owing to errors in visual perception and judgment. Deep learning is a branch of artificial intelligence (AI) that has shown promise to provide real-time intraoperative guidance and augment surgical performance. This study aims to train and test the performance of a deep learning model that can identify inappropriate landing zones on fluoroscopy during endovascular aneurysm repair (EVAR).

New Approach to Generate Ratiometric Signals on Immunochromatographic Strips for Small Molecules.

The self-calibration capability of ratiometric signals has been widely considered to enhance the accuracy, sensitivity, and anti-interference ability of immunoassays. Exploring a new approach to generate ratiometric signals can provide more options for various requirements. Herein, we integrated the negative-readout competitive and positive-readout noncompetitive immunoassays into a single assay by employing different color tracers, labeled peptidomimetic and anti-immunocomplex peptides, to create a new unconstrained ratiometric signal approach. Using an immunochromatographic strip (ICS) and a fungicide benzothiostrobin as the analytical platform and analyte, respectively, we showed that this approach can be extensively applied to fluorescence and colorimetry readouts, which have also been proven for strong anti-interference ability to an external light environment. Moreover, the enormous intuitional color changes of ratiometric fluorescent and colorimetric ICSs (RFICS and RCICS) enabled the formation of the color reference cards (like the pH paper) for visual judgment. After adaptation with a portable smartphone, the quantitative detection limits for RFICS and RCICS were 0.17 and 0.44 ng mL-1, respectively. In addition, the ICSs showed good accuracy for the detection of benzothiostrobin in spiked samples.

A Diffusion Tensor Imaging Study on the White Matter Structures Related to the Phonology in Cantonese-Mandarin Bilinguals.

Cantonese and Mandarin are logographic languages, and the phonology is the main difference between the two languages. It is unclear whether the long-term experience of Cantonese–Mandarin bilingualism will shape different brain white matter structures related to phonological processing. A total of 30 Cantonese–Mandarin bilinguals and 30 Mandarin monolinguals completed diffusion-weighted imaging scan and phonological processing tasks. The tractography and tract-based spatial statistics were used to investigate the structural differences in the bilateral superior longitudinal fasciculus (SLF), inferior longitudinal fasciculus (ILF), and inferior fronto–occipital fasciculus (IFOF) between Cantonese–Mandarin bilinguals and Mandarin monolinguals. The post-hoc correlation analysis was conducted to investigate the relationship between the different structures with phonological processing skills. Compared to the Mandarin monolinguals, the Cantonese–Mandarin bilinguals had higher fractional anisotropy (FA) along the left ILFs higher mean diffusivity (MD) along the right IFOF and the temporoparietal segment of SLF (tSLF), higher axial diffusivity (AD) in the right IFOF and left ILF, and lower number of streamlines in the bilateral tSLF. The mean AD of the different voxels in the right IFOF and the mean FA of the different voxels in the left ILF were positively correlated with the inverse efficiency score (IES) of the Cantonese auditory and Mandarin visual rhyming judgment tasks, respectively, within the bilingual group. The correlation between FA and IES was different among the groups. The long-term experience of Cantonese–Mandarin bilinguals shapes the different brain white matter structures in tSLF, IFOF, and ILF. Compared to the monolinguals, the bilinguals’ white matter showed higher diffusivity, especially in the axonal direction. These changes were related to bilinguals’ phonological processing.

Spatial frequency processing and its modulation by emotional content in severe alcohol use disorder.

Visuo-perceptive deficits in severe alcohol use disorder (SAUD) remain little understood, notably regarding the respective involvement of the two main human visual streams, i.e., magnocellular (MC) and parvocellular (PC) pathways, in these deficits. Besides, in healthy populations, low-level visual perception can adapt depending on the nature of visual cues, among which emotional features, but this MC and PC pathway adaptation to emotional content is unexplored in SAUD. To assess MC and PC functioning as well as their emotional modulations in SAUD. We used sensitivity indices (d') and repeated-measures analyses of variance to compare orientation judgments of Gabor patches sampled at various MC- and PC-related spatial frequencies in 35 individuals with SAUD and 38 matched healthy controls. We then explored how emotional content modulated performances by introducing neutral or fearful face cues immediately before the Gabor patches and added the type of cue in the analyses. SAUD patients showed a general reduction in sensitivity across all spatial frequencies, indicating impoverished processing of both coarse and fine-scale visual content. However, we observed selective impairments depending on facial cues: individuals with SAUD processed intermediate spatial frequencies less efficiently than healthy controls following neutral faces, whereas group differences emerged for the highest spatial frequencies following fearful faces. Altogether, SAUD was associated with mixed MC and PC deficits that may vary according to emotional content, in line with aflexible but suboptimal use of low-level visual content. Such subtle alterations could have implications for everyday life's complex visual judgments.

Application of Multimodal Neuroimaging in the Treatment of Neurological Patients.

Objective In order to study the application of multimodal neuroimaging in the treatment of neurological patients, the brain of patients was scanned and identified through multimodal neuroimaging, so as to provide basis for doctors to judge their diseases and give treatment plans. Method Understand the principle of multimodal neural scene currently used through literature analysis, analyze the current level of neurological diseases in the medical field and the application of multimodal neural image in diseases, recalculate the imaging examination data through neurofuzzy algorithm, and obtain a more optimized identification method than the previous visual judgment method. Result The image inspection data is dimensioned through computer spatial convolution. At this time, the output two-dimensional array is not accurate enough. It needs to be processed again through fuzzy spatial convolution to obtain relatively accurate data output, generate a small two-dimensional array, and use the convolution core for the spatial convolution process of two-dimensional array. The algorithm uses neural network machine learning algorithm to identify and judge the inspection data. Conclusion Through this algorithm to identify the examination data, the early diagnosis sensitivity of intracranial space occupying lesions and intracranial hematogenous lesions are more than 20% higher than the previous traditional recognition methods, which provides a medical imaging basis for the early diagnosis and treatment of acoustic diseases, and improves the treatment probability and prognosis quality of life of neurological patients.

Changing Finger Movement Perception: Influence of Active Haptics on Visual Dominance

The perception of one’s own body is a complex mechanism that can be disturbed by conflicting sensory information and lead to illusory (mis-) perceptions. Prominent models of multisensory integration propose that sensory streams are integrated according to their reliability by approximating Bayesian inference. As such, when considering self-attribution of seen motor actions, previous works argue in favor of visual dominance over other sensations, and internal cues. In the present work, we use virtual reality and a haptic glove to investigate the influence of an active haptic feedback on one’s visual and agency judgments over a performed finger action under experimentally manipulated visual and haptic feedbacks. Data overall confirm that vision dominates for agency judgment in conditions of multisensory conflict. Interestingly, we also show that participants’ visual judgment over their finger action is sensitive to multisensory conflicts (vision, proprioception, motor afferent signals, and haptic perception), thus bringing an important nuance to the widely accepted view on a general visual dominance.

Learning-Replay Based Automated Robotic Testing for Mobile App

Record-replay testing is widely used in mobile app testing as an automated testing method. However, the current record-replay methods are closely dependent on the internal information of the device or app under test. Due to the diversity of mobile devices and system platforms, their practical use is limited. To break this limitation, this paper proposes an entirely black-box learning-replay testing approach by combining robotics and vision technology to achieve a record-replay testing that can support cross-device and cross-platform. Firstly, vision technology is used to extract the critical information of GUI and gesture actions during the tester’s testing process; secondly, the GUI composition and test actions are analyzed to form a test sequence; finally, the robotic arm is guided to complete the replay of the test sequence through visual judgment. On the one hand, the approach in this paper does not access the interior of the app, shielding the association between test actions and device; on the other hand, it captures more abstract test action information instead of simple operation location records and supports more flexible test action replay. We demonstrate the effectiveness of this approach by evaluating the learning-replay of 12 popular apps for 13 typical scenarios on the same device, across devices, and across platforms.

Distinguishing mirror from glass: A "big data" approach to material perception.

Distinguishing mirror from glass is a challenging visual inference, because both materials derive their appearance from their surroundings, yet we rarely experience difficulties in telling them apart. Very few studies have investigated how the visual system distinguishes reflections from refractions and to date, there is no image-computable model that emulates human judgments. Here we sought to develop a deep neural network that reproduces the patterns of visual judgments human observers make. To do this, we trained thousands of convolutional neural networks on more than 750,000 simulated mirror and glass objects, and compared their performance with human judgments, as well as alternative classifiers based on “hand-engineered” image features. For randomly chosen images, all classifiers and humans performed with high accuracy, and therefore correlated highly with one another. However, to assess how similar models are to humans, it is not sufficient to compare accuracy or correlation on random images. A good model should also predict the characteristic errors that humans make. We, therefore, painstakingly assembled a diagnostic image set for which humans make systematic errors, allowing us to isolate signatures of human-like performance. A large-scale, systematic search through feedforward neural architectures revealed that relatively shallow (three-layer) networks predicted human judgments better than any other models we tested. This is the first image-computable model that emulates human errors and succeeds in distinguishing mirror from glass, and hints that mid-level visual processing might be particularly important for the task.

Quantification of SPECT Concentric Ring Artifacts by Radiomics and Radial Features

(1) Background: Concentric ring artifacts in reconstructed SPECT images indicate the presence of detector non-uniformity in gamma camera systems. The identification of these artifacts is generally visual and not quantitative. The aim of our study was to evaluate observer assessments of the presence of concentric rings in reconstructed SPECT phantom images and to verify whether quantitative texture analysis can detect such artifacts, which are detrimental to accurate tumor detection. (2) Methods: Test data were acquired as part of the quarterly quality assurance program using a standardized SPECT phantom containing solid spheres, solid rods, and a water solution of 99mTc. Forty separate SPECT acquisitions were analyzed to assess the presence of ring artifacts. Two experienced medical physicists independently reviewed transaxial images and graded the severity of artifacts on a five-point scale. Quantitative radiomic features were computed for volumes of interest located in the uniform phantom section. In addition to these, radial contrast (RContrast) and radial root-mean-square contrast (RRMSC) were also calculated and derived from the radial profile of summed slices transformed into polar coordinates. (3) Results: Artifacts were considered sufficiently severe to warrant camera re-tuning in 10 rod sections, 17 sphere sections, and 16 uniform sections. In the uniform sections, there was “good agreement” for inter-observer and intra-rater assessments (κ = 0.66, Fisher exact p < 0.0001 and κ = 0.61, and Fisher exact p = 0.001, respectively). The two radial features agreed significantly (p < 0.001) with visual severity judgment of ring artifacts in uniform sections and were selected as informative about the presence of ring artifacts by LASSO approach. The increased magnitude of RContrast and RRMSC correlated significantly with increasingly severe artifact scores (ρ = 0.65–0.66, p < 0.0001). (4) Conclusions: There was good agreement between the physicists with respect to the presence of circular ring artifacts in uniform sections of SPECT quality assurance scans, with the artifacts accurately detected by radial contrast and noise-to-signal ratio measurements.

Multi-layer perceptron for detection of different class antibiotics from visual fluorescence response of a carbon nanoparticle-based multichannel array sensor

Lack of automated accurate decision-making along with an on-site detection system impedes the identification of substances of environmental concern. In pursuit of making this feasible, we interconnected our optical fluorescence array sensing strategy with the predictive analytics of artificial intelligence. Herein, we developed a Carbon Nanoparticle-based nine-channel fluorescence array sensing method for the detection of six antibiotics of different classes that are precariously dumped in the environment from various industrial and animal husbandry sources. The fluorescence responses of the arrays in the presence or absence of six antibiotics were captured digitally and these were utilized as feature values for the identification of classes using machine learning and deep learning algorithms. Among the seven tested multi-class classification algorithms, Multi-layer Perceptron (MLP) with Generative Adversarial Nets stimulated augmented data set (Aug-MLP) outdid the others in recognizing the antibiotics. Most importantly, the performance of Aug-MLP is comparable to fluorescence spectroscopic discrimination that outclasses human visual judgment. The whole methodology was found to adapt well in real samples like extracts of poultry feeds. In a nutshell, a nanotechnology-deep learning interfaced semi-automated on-site multi-class antibiotic detection strategy has been developed that could be extended for inexpensive and expedited detection of other chemical entities.

Lateralization as a symphony: Joint influence of interhemispheric inhibition and transmission on brain asymmetry and syntactic processing

This study investigated the roles of cross-hemispheric communications in promoting left-lateralized syntactic processing in the brain. Fifty-six young right-handers without familial sinistrality background underwent a divided visual field ERP grammaticality judgment experiment to assess syntactic processing in each hemisphere. Two behavioral tasks —the bilateral flanker task and bilateral word matching task, were used to assess cross-hemispheric inhibition and transmission. Grand average ERP data showed a significant P600 grammaticality effect in the left hemisphere (LH) only; however, individual variations in the P600 responses were observed in both hemispheres. Results of correlational analyses showed that larger LH P600 effects were associated with slower inter-hemispheric transmissions; smaller right hemisphere (RH) P600 effects were associated with more effective RH inhibition. These results yielded support for both the callosal distance hypothesis and the inhibition hypothesis for language lateralization and demonstrated that different aspects of cross-hemispheric communications jointly influence the degree of syntactic lateralization.

Surface Defect Detection Method Based on Improved Attention Mechanism and Feature Fusion Model.

Cylinder liners are important to automobile engines. The appearance quality will directly affect the life and safety of the engines. At present, the appearance quality inspection of cylinder liners mainly relies on manual visual judgment, which is easily affected by the subjective factors of inspectors. This paper studies improved machine vision to realize surface defect detection. It proposes the improvement of the attention mechanism and a feature fusion method to locate and classify the defect. Experiments show that the method proposed in this paper has improved both accuracy and speed, and it can detect defects in production and realize industrialization. At the same time, the method studied in this paper has the value of popularization and application for appearance defect detection in other fields.

Impact of ventricle size on neuropsychological outcomes in treated pediatric hydrocephalus: an HCRN prospective cohort study.

In pediatric hydrocephalus, shunts tend to result in smaller postoperative ventricles compared with those following an endoscopic third ventriculostomy (ETV). The impact of the final treated ventricle size on neuropsychological and quality-of-life outcomes is currently undetermined. Therefore, the authors sought to ascertain whether treated ventricle size is associated with neurocognitive and academic outcomes postoperatively. This prospective cohort study included children aged 5 years and older at the first diagnosis of hydrocephalus at 8 Hydrocephalus Clinical Research Network sites from 2011 to 2015. The treated ventricle size, as measured by the frontal and occipital horn ratio (FOR), was compared with 25 neuropsychological tests 6 months postoperatively after adjusting for age, hydrocephalus etiology, and treatment type (ETV vs shunt). Pre- and posttreatment grade point average (GPA), quality-of-life measures (Hydrocephalus Outcome Questionnaire [HOQ]), and a truncated preoperative neuropsychological battery were also compared with the FOR. Overall, 60 children were included with a mean age of 10.8 years; 17% had ≥ 1 comorbidity. Etiologies for hydrocephalus were midbrain lesions (37%), aqueductal stenosis (22%), posterior fossa tumors (13%), and supratentorial tumors (12%). ETV (78%) was more commonly used than shunting (22%). Of the 25 neuropsychological tests, including full-scale IQ (q = 0.77), 23 tests showed no univariable association with postoperative ventricle size. Verbal learning delayed recall (p = 0.006, q = 0.118) and visual spatial judgment (p = 0.006, q = 0.118) were negatively associated with larger ventricles and remained significant after multivariate adjustment for age, etiology, and procedure type. However, neither delayed verbal learning (p = 0.40) nor visual spatial judgment (p = 0.22) was associated with ventricle size change with surgery. No associations were found between postoperative ventricle size and either GPA or the HOQ. Minimal associations were found between the treated ventricle size and neuropsychological, academic, or quality-of-life outcomes for pediatric patients in this comprehensive, multicenter study that encompassed heterogeneous hydrocephalus etiologies.

Reimagining Imagia. Neurocognitive Theory of Image Revisited With the Consciousness Studies Perspective

Imagia. Towards a neurocognitive theory of image (Francuz, 2013) defines art perception as a synthesis between acts of perception and visual imagination limited by attentional constraints. The majority of empirical evidence the author referred to in the book focuses on visual processing, attention, and aesthetic judgements. Nevertheless, other mechanisms involved in the subjective experience formation undoubtedly also plays a role in the perception of art. Thus, it seems worth complementing the fundaments of the neurocognitive theory of image with consciousness literature. The research in this domain often reminds us that we see much more than is presented. Our conscious perception is modified, extended, and enriched due to multiple mechanisms, such as prediction, integration, or learning. Here, I review selected examples of work from the consciousness literature to illustrate this claim. Furthermore, I discuss this work in the context of art perception. Finally, I reconstruct the model proposed by Francuz and offer its extension that accounts for the aspects covered by the original work and those addressed in this paper.

Dissecting Breeders' Sense via Explainable Machine Learning Approach: Application to Fruit Peelability and Hardness in Citrus.

“Genomics-assisted breeding”, which utilizes genomics-based methods, e.g., genome-wide association study (GWAS) and genomic selection (GS), has been attracting attention, especially in the field of fruit breeding. Low-cost genotyping technologies that support genome-assisted breeding have already been established. However, efficient collection of large amounts of high-quality phenotypic data is essential for the success of such breeding. Most of the fruit quality traits have been sensorily and visually evaluated by professional breeders. However, the fruit morphological features that serve as the basis for such sensory and visual judgments are unclear. This makes it difficult to collect efficient phenotypic data on fruit quality traits using image analysis. In this study, we developed a method to automatically measure the morphological features of citrus fruits by the image analysis of cross-sectional images of citrus fruits. We applied explainable machine learning methods and Bayesian networks to determine the relationship between fruit morphological features and two sensorily evaluated fruit quality traits: easiness of peeling (Peeling) and fruit hardness (FruH). In each of all the methods applied in this study, the degradation area of the central core of the fruit was significantly and directly associated with both Peeling and FruH, while the seed area was significantly and directly related to FruH alone. The degradation area of albedo and the area of flavedo were also significantly and directly related to Peeling and FruH, respectively, except in one or two methods. These results suggest that an approach that combines explainable machine learning methods, Bayesian networks, and image analysis can be effective in dissecting the experienced sense of a breeder. In breeding programs, collecting fruit images and efficiently measuring and documenting fruit morphological features that are related to fruit quality traits may increase the size of data for the analysis and improvement of the accuracy of GWAS and GS on the quality traits of the citrus fruits.

Artificial neural network approach for multiphase segmentation of battery electrode nano-CT images

The segmentation of tomographic images of the battery electrode is a crucial processing step, which will have an additional impact on the results of material characterization and electrochemical simulation. However, manually labeling X-ray CT images (XCT) is time-consuming, and these XCT images are generally difficult to segment with histographical methods. We propose a deep learning approach with an asymmetrical depth encode-decoder convolutional neural network (CNN) for real-world battery material datasets. This network achieves high accuracy while requiring small amounts of labeled data and predicts a volume of billions voxel within few minutes. While applying supervised machine learning for segmenting real-world data, the ground truth is often absent. The results of segmentation are usually qualitatively justified by visual judgement. We try to unravel this fuzzy definition of segmentation quality by identifying the uncertainty due to the human bias diluted in the training data. Further CNN trainings using synthetic data show quantitative impact of such uncertainty on the determination of material’s properties. Nano-XCT datasets of various battery materials have been successfully segmented by training this neural network from scratch. We will also show that applying the transfer learning, which consists of reusing a well-trained network, can improve the accuracy of a similar dataset.

A Novel Object Detection Model Based on Faster R-CNN for Spodoptera frugiperda According to Feeding Trace of Corn Leaves

The conventional method for crop insect detection based on visual judgment of the field is time-consuming, laborious, subjective, and error prone. The early detection and accurate localization of agricultural insect pests can significantly improve the effectiveness of pest control as well as reduce the costs, which has become an urgent demand for crop production. Maize Spodoptera frugiperda is a migratory agricultural pest that has severely decreased the yield of maize, rice, and other kinds of crops worldwide. To monitor the occurrences of maize Spodoptera frugiperda in a timely manner, an end-to-end Spodoptera frugiperda detection model termed the Pest Region-CNN (Pest R-CNN) was proposed based on the Faster Region-CNN (Faster R-CNN) model. Pest R-CNN was carried out according to the feeding traces of maize leaves by Spodoptera frugiperda. The proposed model was trained and validated using high-spatial-resolution red–green–blue (RGB) ortho-images acquired by an unmanned aerial vehicle (UAV). On the basis of the severity of feeding, the degree of Spodoptera frugiperda invasion severity was classified into the four classes of juvenile, minor, moderate, and severe. The degree of severity and specific feed location of S. frugiperda infestation can be determined and depicted in the frame forms using the proposed model. A mean average precision (mAP) of 43.6% was achieved by the proposed model on the test dataset, showing the great potential of deep learning object detection in pest monitoring. Compared with the Faster R-CNN and YOLOv5 model, the detection accuracy of the proposed model increased by 12% and 19%, respectively. Further ablation studies showed the effectives of channel and spatial attention, group convolution, deformable convolution, and the multi-scale aggregation strategy in the aspect of improving the accuracy of detection. The design methods of the object detection architecture could provide reference for other research. This is the first step in applying deep-learning object detection to S. frugiperda feeding trace, enabling the application of high-spatial-resolution RGB images obtained by UAVs to S. frugiperda-infested object detection. The proposed model will be beneficial with respect to S. frugiperda pest stress monitoring to realize precision pest control.