Difficulties In Recognition Research Articles

EXPRESS: Alexithymia does not explain facial expression recognition difficulties across the dark triad spectrum.

Using an individual differences approach (N=236) we assessed whether trait alexithymia was able to predict unique variance in facial expression discrimination ability and facial expression labelling ability above and beyond an individual's level of dark triad traits. Alexithymia has been shown to be associated with the Dark Triad, and recent evidence has suggested that co-occurring alexithymia may explain facial expression recognition difficulties in autism spectrum disorder. Here we tested this alexithymia hypothesis for individuals on the dark triad spectrum. Results showed that autistic traits, alexithymic traits and dark triad traits all correlated with expression recognition ability. However, linear regression models showed that an individual's level of dark triad traits, their level of autistic traits, and a brief measure of general cognitive ability each predicted unique variance in facial expression discrimination and facial expression labelling ability, but an individual's level of alexithymic traits predicted no additional unique variance. We suggest that Dark Triad and Autistic traits each uniquely contribute to expression recognition ability alongside general cognitive ability.

Align, enhance and read: Scene Tibetan text recognition with cross-sequence reasoning

Market demand has primarily driven scene text recognition to focus on widely spoken languages, such as English and Chinese. Current research put scarce attention on Tibetan, despite its substantial demand and potential applications. To fill this gap, we analyze the characteristics of Tibetan text and point out the corresponding difficulties in text recognition. Moreover, this study explores the performance degradation of existing recognition methods on Tibetan text, and proposes a flexible alignment decoding manner with cross-modal sequence reasoning for Tibetan text. Concretely, our work leverages the information between different modal sequences to learn detailed visual features for Tibetan characters with complex glyphs through cross-modal sequence reasoning. Furthermore, we design an innovative alignment decoding scheme that aligns characters in a dense manner, improving attention quality for the character alignment. Experimental results indicate that the performance of existing text recognition on Tibetan text is far behind that of English or Chinese text. With our design, the recognition on Tibetan text achieves significant improvement.

Hiding in plain sight: children with visual perceptual difficulties in schools

Cerebral visual impairment (CVI) is increasingly being recognized as a significant cause of visual difficulties in children, particularly those with typical visual acuity, who nonetheless struggle in educational settings. This narrative review aims to elucidate the nature and impact of visual perceptual difficulties (VPD) associated with CVI in school-aged children, who often remain undiagnosed due to the current erroneous focus on visual acuity as a required diagnostic criterion for CVI. The review synthesizes findings from recent studies, highlighting that up to 3.4% of children in mainstream schools and a higher percentage in special educational settings may experience VPD, which significantly impacts upon their learning and development. The manifestations of VPDs, such as difficulties in motion perception, recognition, and visuospatial processing, are often subtle and can thus be overlooked, leading to misconceptions about the origins of the affected child’s abilities and behaviors. The review also discusses the challenges in current diagnostic processes, emphasizing the need for comprehensive history taking and assessments that go beyond standard visual acuity tests. It proposes a multi-faceted approach to identification and support, incorporating both clinical assessments and teacher/parental observations, to better address the needs of affected children. Furthermore, this paper advocates for the inclusion of VPDs in the International Classification of Diseases (ICD 11) to ensure children with these visual issues receive appropriate educational support. By integrating lived experiences of individuals with CVI and the latest research findings, this review underscores the urgent need for awareness and tailored educational strategies designed to support children with VPDs. The findings suggest that without such recognition and intervention, many children with VPDs will continue to “hide in plain sight,” facing unnecessary challenges in their educational and social development. The review concludes with recommendations for policy changes and future research directions to improve the identification, classification, and support of children with VPDs within the educational system.

Development of a code-switched Hindi-Marathi dataset and transformer-based architecture for enhanced speech recognition using dynamic switching algorithms

Especially in low-resource languages like Hindi and Marathi, code-switching presents major difficulties for automatic speech recognition (ASR). This work provides a 450-hour annotated dataset of Hindi-Marathi code-switching, including tag-switching, intra-sentential, and inter-sentential patterns. We augment a transformer-based ASR architecture with dynamic switching algorithms using wav2vec2 and a reinforcement learning-based approach known as Q-Learning, thereby dynamically optimizing language transition points.With a Word Error Rate (WER) of 0.2800 and a Character Error Rate (CER) of 0.2400, the proposed model beats conventional HMM-GMM and RNN-based ASR systems. Combining reinforcement learning for dynamic code-switching with transformer-based self-supervised learning demonstrates enhanced accuracy and flexibility.Comparative analysis shows the improvements relative to heuristic methods, Kaldi baselines, and pre-trained monolingual models. This work underscores the significance of hybrid architectures, dynamic algorithms, and sophisticated acoustic modeling in code-switched speech recognition, thereby offering a comprehensive framework for multilingual automatic speech recognition. The results have a major impact on the evolution of ASR in linguistically diverse and economically constrained environments.

Basketball robot object detection and distance measurement based on ROS and IBN-YOLOv5s algorithms.

With the combination of artificial intelligence and robotics technology, more and more professional robots are entering the public eye. Basketball robot competition, as a very good target system for autonomous robot research, is very suitable for conducting research on robot autonomous perception system object detection. However, traditional basketball robots have problems such as recognition difficulties, which seriously affect the recognition of robot targets and distance measurement based on recognition. To improve the performance of basketball robots in competitions, research was conducted to improve the object detection system. Firstly, a basketball robot object detection system based on robot operating system was designed. In the software layer of the object detection system, an algorithm that combines YOLOv5s and laser detection was used, and an appropriate instance batch normalization network module was introduced in the YOLOv5s algorithm to improve the model's generalization ability. The experiment outcomes indicated that the improved algorithm had intersection over union (IoU), structural information loss, ambiguity and signal-to-noise ratio of 0.96, 0.03, 0.13, and 0.98, respectively, and performed the best in the other comparison models. The recall curve area and F1 value of the improved algorithm were 0.95 and 0.9789, respectively. In the detection of basketball, volleyball, and calibration columns, the average classification accuracy of the improved model was 95.87%, and the average calibration box accuracy was 97.05%. From this, the algorithm proposed in the study has robust performance and can efficiently achieve object detection and recognition of basketball robots. The improved algorithm proposed in the study provides more reliable and rich information for the perception ability of basketball robots, as well as for their subsequent decision-making and action planning, thereby improving the overall technical level of the robots.

Beyond body image: Self-face recognition and negative self-face evaluations in women at high risk for an eating disorder.

Although one's face represents a core aspect of one's physical appearance, it remains underexplored in eating disorder (ED) populations. The current study examined whether self-face recognition accuracy and evaluations differ in those at high (vs. low) ED risk. We also explored whether recognition accuracy and evaluations are related, and whether this relationship is moderated by ED risk status. Participants included 222 women of East/Southeast Asian (61%) and Caucasian (39%) ethnicity, stratified into low-ED-risk (n=124) and high-ED-risk (n=98). Participants completed measures of self-face recognition accuracy (difficulty recognizing their face and sensitivity to seeing their facial image) and evaluations (including perceived facial adiposity and attractiveness, and head dissatisfaction). Partially consistent with hypotheses, regression analyses revealed that participants at high (vs. low) risk showed poorer self-face recognition accuracy (greater difficulty) and more negative self-face evaluations (higher perceived adiposity, lower perceived attractiveness, greater head dissatisfaction). However, ED risk was not significantly associated with self-face recognition sensitivity. Inconsistent with expectations, whilst greater self-face recognition difficulty was associated with more negative self-face evaluations, ED risk failed to moderate such effects. Findings suggest the potential value of considering face and body image disturbances in ED research and interventions. Further research is encouraged.

Automatic Defects Recognition of Lap Joint of Unequal Thickness Based on X-Ray Image Processing.

It is difficult to automatically recognize defects using digital image processing methods in X-ray radiographs of lap joints made from plates of unequal thickness. The continuous change in the wall thickness of the lap joint workpiece causes very different gray levels in an X-ray background image. Furthermore, due to the shape and fixturing of the workpiece, the distribution of the weld seam in the radiograph is not vertical which results in an angle between the weld seam and the vertical direction. This makes automatic defect detection and localization difficult. In this paper, a method of X-ray image correction based on invariant moments is presented to solve the problem. In addition, a novel background removal method based on image processing is introduced to reduce the difficulty of defect recognition caused by variations in grayscale. At the same time, an automatic defect detection method combining image noise suppression, image segmentation, and mathematical morphology is adopted. The results show that the proposed method can effectively recognize the gas pores in an automatic welded lap joint of unequal thickness, making it suitable for automatic detection.

The Applicability and Reflection Characteristics of Coal Failure Events for External Monitoring-While-Drilling of Underground Pressure Relief Drilling

Previous research results preliminarily indicated that the Coal Failure Events (CFEs) that occurred during the process of Underground Pressure Relief Drilling (UPRD) represented the phenomenon of coal fracture and energy release. The research results had excellent value for the monitoring and response of pressure relief drilling while drilling, but there were still some special situations that needed to be analyzed and studied in actual on-site testing. So, through on-site testing and data statistical analysis, the study investigated the applicability of the innovative external Monitoring-While-Drilling (MWD) method for UPRD with more coal failure events and made a quantitative statistic of the CFEs and their relationship with abutment pressure to reveal the applicability of the external MWD method and characteristic of CFEs. The results showed that hundreds of CFEs were produced in the UPRD process, which must be removed to ensure the accuracy of the MWD method. Although CFEs bring recognition difficulties, they also provide conditions for studying their own distribution and characteristics. Results showed that more CFEs were produced in the depth of difficult drilling, which indicated that there was a positive correlation between the degree of difficulty in drilling and the number of CFEs. In addition, spectrum analysis showed that the depths with more CFE occurrence were more likely to produce high-frequency events. When the surrounding stress of drilling rocks is high, the occurrence of small fractures with a higher main frequency may become more frequent and consistent; more fractures with similar failure forms would occur, which may have a lower fractal dimension and promote the generation of more failure. The research results were of great significance for the MWD method for UPRD, a quantitative study of CFEs and their generation characteristics during UPRD construction.

Global changes in the pattern of connectivity in developmental prosopagnosia.

Developmental prosopagnosia is a neurodevelopmental condition characterized by difficulties in recognizing the identity of a person from their face. While current theories of the neural basis of developmental prosopagnosia focus on the face processing network, successful recognition of face identities requires broader integration of neural signals across the whole brain. Here, we asked whether disruptions in global functional and structural connectivity contribute to the face recognition difficulties observed in developmental prosopagnosia. We found that the left temporal pole was less functionally connected to the rest of the brain in developmental prosopagnosia. This was driven by weaker contralateral connections to the middle and inferior temporal gyri, as well as to the medial prefrontal cortex. The pattern of global connectivity in the left temporal pole was also disrupted in developmental prosopagnosia. Critically, these changes in global functional connectivity were only evident when participants viewed faces. Structural connectivity analysis revealed localized reductions in connectivity between the left temporal pole and a number of regions, including the fusiform gyrus, inferior temporal gyrus, and orbitofrontal cortex. Our findings underscore the importance of whole-brain integration in supporting typical face recognition and provide evidence that disruptions in connectivity involving the left temporal pole may underlie the characteristic difficulties of developmental prosopagnosia.

Real Time Object Detection Algorithm in Foggy Weather Based on WVIT-YOLO Model

Introduction: To address the challenges of low visibility, object recognition difficulties, and low detection accuracy in foggy weather, this paper introduces the WViT-YOLO real-time fog detection model, built on the YOLOv5 framework. The NVIT-Net backbone network, incorporating NTB and NCB modules, enhances the model's ability to extract both global and local features from images. Method: An efficient convolutional C3_DSConv module is designed and integrated with channel attention mechanisms and ShuffleAttention at each upsampling stage, improving the model's computational speed and its ability to detect small and blurry objects. The Wise-IOU loss function is utilized during the prediction stage to enhance the model's convergence efficiency. Result: Experimental results on the publicly available RTTS dataset for vehicle detection in foggy conditions demonstrate that the WViT-YOLO model achieves a 3.2% increase in precision, a 9.5% rise in recall, and an 8.6% improvement in mAP50 compared to the baseline model. Furthermore, WViT-YOLO shows a 9.5% and 8.6% mAP50 improvement over YOLOv7 and YOLOv8, respectively. For detecting small and blurry objects in foggy conditions, the model demonstrates approximately a 5% improvement over the benchmark, significantly enhancing the detection network's generalization ability under foggy conditions. Conclusion: This advancement is crucial for improving vehicle safety in such weather. The code is available at https://github.com/QinghuaZhang1/mode.

Poor face recognition predicts social anxiety in autism: A short report.

Research has shown that some autistic people have severe difficulties in recognising other people's faces. However, little is understood about how these difficulties impact the daily life and the mental well-being of autistic people. In this study, we asked 60 autistic adults with varying degrees of face recognition ability to complete two tests of face recognition, a questionnaire about social anxiety and a bespoke survey which asked participants about their experiences of face recognition and social interaction. We found that participants who had poor face recognition reported experiencing higher levels of social anxiety compared to those with average or better face recognition skills. More than half felt that their face recognition difficulties affected their social interactions, and over a third believed it hindered their ability to make friends. These findings suggest that face recognition difficulties may contribute to social anxiety among autistic individuals.

지방 중·소도시 지역아동센터 아동들의 어휘 및 읽기 발달 현황

Purpose: This study aimed to examine the receptive and expressive vocabulary and reading achievement of children attending community children's centers in small and medium-sized cities to identify the current status of vocabulary and reading development of children attending community children's centers. Method: Vocabulary and reading achievement tests were conducted on 82 children attending four community children's centers in A City, a medium-sized city. Based on the test results, receptive and expressive vocabulary and reading achievement status were analyzed, the types of reading disorders of children identified as reading disabled were analyzed, and the distribution of children with overlapping vocabulary delay and reading disorders was investigated. Results: First, 41.4% of all children showed a delay in expressive vocabulary, and 52.4% showed a delay in receptive vocabulary. Second, 55.9% of all children showed significantly low achievement in the reading achievement test. Third, among the children with reading disability, 80% of the lower grades had difficulty in word recognition. On the other hand, 55.6% of the upper grades had difficulty in word recognition, and 38.9% of the children showed difficulty in reading fluency. Fourth, among the children with vocabulary delay or reading disability, 66.7% of the children showed overlapping vocabulary delay and reading disability. Conclusion: Because a high rate of children in community children's centers shows vocabulary delay and reading difficulties, it was suggested that language and reading support should be provided to children using community children's centers, and administrative and financial support is needed for this.

Multi-source domain adaptation for EEG emotion recognition based on inter-domain sample hybridization.

Electroencephalogram (EEG) is widely used in emotion recognition due to its precision and reliability. However, the nonstationarity of EEG signals causes significant differences between individuals or sessions, making it challenging to construct a robust model. Recently, domain adaptation (DA) methods have shown excellent results in cross-subject EEG emotion recognition by aligning marginal distributions. Nevertheless, these methods do not consider emotion category labels, which can lead to label confusion during alignment. Our study aims to alleviate this problem by promoting conditional distribution alignment during domain adaptation to improve cross-subject and cross-session emotion recognition performance. This study introduces a multi-source domain adaptation common-branch network for EEG emotion recognition and proposes a novel sample hybridization method. This method enables the introduction of target domain data information by directionally hybridizing source and target domain samples without increasing the overall sample size, thereby enhancing the effectiveness of conditional distribution alignment in domain adaptation. Cross-subject and cross-session experiments were conducted on two publicly available datasets, SEED and SEED-IV, to validate the proposed model. In cross-subject emotion recognition, our method achieved an average accuracy of 90.27% on the SEED dataset, with eight out of 15 subjects attaining a recognition accuracy higher than 90%. For the SEED-IV dataset, the recognition accuracy also reached 73.21%. Additionally, in the cross-session experiment, we sequentially used two out of the three session data as source domains and the remaining session as the target domain for emotion recognition. The proposed model yielded average accuracies of 94.16 and 75.05% on the two datasets, respectively. Our proposed method aims to alleviate the difficulties of emotion recognition from the limited generalization ability of EEG features across subjects and sessions. Though adapting the multi-source domain adaptation and the sample hybridization method, the proposed method can effectively transfer the emotion-related knowledge of known subjects and achieve accurate emotion recognition on unlabeled subjects.

A semantic segmentation network based on CNN and Swin transformer for landslide detection

Abstract To address the issues of poor object localization, difficulty in object recognition, and inadequate segmentation performance of CNN-based semantic segmentation methods in landslide detection, a semantic segmentation model based on the Swin transformer and ResUNet architectures for landslide detection in remote sensing images is proposed in this paper. This method combines the global feature capabilities of the Swin transformer with the local feature extraction abilities of CNN and integrates our proposed RCCT module for landslide segmentation in remote sensing images. We apply this model to the identification and extraction of landslide disasters in Menyuan County and the Sanjiangyuan region of Qinghai Province, China, achieving a promising F1 Score of 81.91. Furthermore, we conducted a comprehensive evaluation of our proposed model on the BiJie Landslide dataset in comparison to four recently developed methods for landslide recognition. The experimental findings indicate the superior performance and effectiveness of our model over the comparative approaches.

Visual Target Interaction Modeling with Multichannel Data Fusion

Facing the demand for accurate, fast, and natural human–computer interaction in the multi-screen control environment, the traditional exchange method of frequent switching control of multi-screen through mouse, keyboard, and other manual methods has problems of low efficiency, weak flexibility, and poor experience. In this paper, we research multi-screen precision control technology based on eye movement to break through the difficulties of precise recognition of human vision under complex environments and poor stability of vision estimation under frequent switching control of multi-screen. This study explores the technology for controlling multiple screens with high precision by tracking eye movements. It overcomes the challenges associated with reliably discerning human visual focus in intricate settings and the instability of visual assessments during rapid transitions between multiple screens. The experimental results of 1143 eye-movement capture tasks are studied, including the accurate recognition technology of human head posture and pupil gaze direction under complex backgrounds, illumination, and different visual field environments, and the theoretical modeling method of multi-screen precise target interaction based on eye-movement is explored. The expression data will be processed using convolutional neural networks, and the dataset will be trained through Python programming and tested on test samples collected by simulated flight crews in the laboratory with an accuracy rate of more than 85%.

A new test for evaluation of marginal cognitive function deficits in idiopathic normal pressure hydrocephalus through expressing texture recognition by sound symbolic words.

The number of dementia patients is increasing with population aging. Preclinical detection of dementia in patients is essential for access to adequate treatment. In previous studies, dementia patients showed texture recognition difficulties. Onomatopoeia or sound symbolic words (SSW) are intuitively associated with texture impressions and are less likely to be affected by aphasia and description of material perception can be easily obtained. In this study, we aimed to create a test of texture recognition ability expressed by SSW to detect the presence of mild cognitive disorders. The sound symbolic words texture recognition test (SSWTRT) is constructed from 12 close-up photos of various materials and participants were to choose the best SSW out of 8 choices to describe surface texture in the images in Japanese. All 102 participants seen in Juntendo University Hospital from January to August 2023 had a diagnosis of possible iNPH (age mean 77.9, SD 6.7). The answers were scored on a comprehensive scale of 0 to 1. Neuropsychological assessments included MMSE, FAB, and the Rey Auditory Verbal Learning Test (RAVLT), Pegboard Test, and Stroop Test from the EU-iNPH Grading Scale (GS). In study 1 the correlation between SSWTRT and the neuropsychological tests were analyzed. In study 2, participants were divided into two groups: the Normal Cognition group (Group A, n = 37) with MMSE scores of 28 points or above, and the Mild Cognitive Impairment group (Group B, n = 50) with scores ranging from 22 to 27 points, and its predictability were analyzed. In study 1, the total SSWTRT score had a moderate correlation with the neuropsychological test results. In study 2, there were significant differences in the SSWTRT scores between groups A and B. ROC analysis results showed that the SSWTR test was able to predict the difference between the normal and mildly impaired cognition groups. The developed SSWTRT reflects the assessment results of neuropsychological tests in cognitive deterioration and was able to detect early cognitive deficits. This test not only relates to visual perception but is likely to have an association with verbal fluency and memory ability, which are frontal lobe functions.

Feature balanced re-enhanced network with multi-factor margin loss for long-tailed visual recognition

Real-world data often exhibits a long-tailed distribution, where the number of training samples for head classes far exceeds that of tail classes. This class imbalance phenomenon poses significant challenges for training deep neural networks. Existing class-aware loss methods typically focus only on the numerical relationship between class samples, blindly favoring the optimization of tail classes during the process, while neglecting the difficulty of samples and the similarity between the current class and other classes. To this end, relying only on the number relationship can easily lead to over-fitting of tail classes, thereby failing to fully utilize the potential information in the data. Therefore, we propose the Multi-Factor Margin Loss (MFMLoss), which consists of positive margin loss and negative margin loss. MFMLoss comprehensively considers three factors at three levels: overall, class, and sample: (1) quantitative relationships, (2) inter-class similarity relationships, and (3) sample recognition difficulty. The combined consideration of these three factors enables the model to pay more attention to confusing classes and difficult samples during the training process, rather than solely on tail classes, thus achieving optimization from coarse-grained to fine-grained. To further mitigate the negative impact of the imbalance between head and tail classes on feature learning, we design a new network architecture, called F-BREN. F-BREN consists of two components: the feature balancing network and the feature re-enhancement network. The former is trained with negative margin loss, which reduces the recognizability of easy samples. The latter is trained with positive margin loss, using positive margin to give more attention to hard samples, thus balancing the model’s attention to all samples. We conducted extensive experiments on four long-tailed benchmark datasets: CIFAR10-LT, CIFAR100-LT, ImageNet-LT and iNaturalist 2018, comparing the recognition accuracy of our method with eight state-of-the-art methods. The experimental results demonstrate that our proposed method outperforms the eight compared methods.

Use of a radiology tool for the diagnosis of pulmonary fibrosis

ObjectiveThe purpose of this paper was to perform an exploratory reader study to assess the utility of a web-based application in assisting non-chest radiologist in correctly diagnosing the radiographic pattern of pulmonary fibrosis. MethodsThree non-chest radiologists with 5 to 20 years of experience individually reviewed 3 rounds of randomly chosen chest CT scans (round 1: 100 scans, round 2: 50 scans, round 3: 25 scans) from a list of patients with established diagnosis of pulmonary fibrosis. In round 1, radiologists were asked to directly record their diagnosis for the pattern of fibrosis. In round 2 and 3 they were asked to review for features provided in a web-based application and provide diagnosis based on the most likely predicted diagnosis from the application. There was an approximate 1-month interval and relevant tutorials were provided between each round. Diagnosis accuracy is reported by readers at each round. ResultsThe overall accuracy increased from 63 % (n = 188/299) in round 1 to 74 % in round 3 (n = 52/70) (p = 0.0265). Difficulty in recognition of mosaic attenuation and homogeneous has led to misdiagnosis. Refining the definition for feature homogeneous increased the diagnosis accuracy of NSIP from 42 % (n = 20/48) in round 2 to 65 % (n = 24/37) in round 3(p = 0.0179). The Fleiss Kappa across readers varied from Round 1 to Round 3 with values 0.36 to 0.42. ConclusionsUsing the web-based application with refined definition for feature homogeneous helps to improve the non-subspecialty radiologist's accuracy in diagnosing different types of fibrosis.

A Fault Diagnosis Method for Electric Check Valve Based on ResNet-ELM with Adaptive Focal Loss

Under the trend of carbon neutrality, the adoption of electric mineral transportation equipment is steadily increasing. Accurate monitoring of the operational status of electric check valves in diaphragm pumps is crucial for ensuring transportation safety. However, accurately identifying the operational characteristics of electric check valves under complex excitation and noisy environments remains challenging. This paper proposes a monitoring method for the status of electric check valves based on the integration of Adaptive Focal Loss (AFL) with residual networks and Extreme Learning Machines (AFL-ResNet-ELMs). Firstly, to address the issue of unclear feature representation in one-dimensional vibration signals, grayscale operations are employed to transform the one-dimensional data into grayscale images with more distinct features. Residual networks are then utilized to extract the state features of the check valve, with Extreme Learning Machines serving as the feature classifier. Secondly, to overcome the issue of imbalanced industrial data distribution, a new Adaptive Focal Loss function is designed. This function focuses the training process on difficult-to-classify data samples, balancing the recognition difficulty across different samples. Finally, experimental studies are conducted using industrially measured vibration data of the electric check valve. The results indicate that the proposed method achieves an average accuracy of 99.60% in identifying four health states of the check valve. This method provides a novel approach for the safety monitoring of slurry pipeline transportation processes.

Citrus recognition in orchard scene based on modified HSV-morphology method

The orchard has always been an important scene for citrus pickers, and the existence of factors such as leaf occlusion and color similarity always lead to the difficulty of robot recognition. Based on a citrus image dataset collected from actual orchard scenes, we extract image features and build a mathematical model to identify and count the number of oranges in each image, and show the distribution of apples in the entire dataset. Firstly, this paper establishes a model based on HSV method, focusing on recognizing yellow citrus. Secondly, with the further analysis of the data set, it is found that the cyan citrus exists. Therefore, this paper defines the empirical HSV range of two kinds of citrus at the same time and introduces contour processing to optimize the model. In order to accurately distinguish the immature citrus from the leaves with similar colors and the citrus occluded by leaves, this paper uses more detailed color threshold setting and morphological operation, and makes the eccentricity of the contour less than 0.85. At the same time, considering the size and overlap of citrus, morphological operations such as dilation and erosion and contour detection techniques are applied to further improve the recognition ability of the model. Finally, the experimental results show that the average accuracy of the established model reaches 92.1%, which has a good recognition effect on citrus.