Deep learning methods have been successfully applied in medical image classification, segmentation and detection tasks. The U-Net architecture has been widely applied for these tasks. In this paper, we propose a U-Net variant for improved vessel segmentation in retinal fundus images. Firstly, we design a minimal U-Net (Mi-UNet) architecture, which drastically reduces the parameter count to 0.07M compared to 31.03M for the conventional U-Net. Moreover, based on Mi-UNet, we propose Salient U-Net (S-UNet), a bridge-style U-Net architecture with a saliency mechanism and with only 0.21M parameters. S-UNet uses a cascading technique that employs the foreground features of one net block as the foreground attention information of the next net block. This cascading leads to enhanced input images, inheritance of the learning experience of previous net blocks, and hence effective solution of the data imbalance problem. S-UNet was tested on two benchmark datasets, DRIVE and CHASE_DB1, with image sizes of 584 × 565 and 960 × 999, respectively. S-UNet was tested on the TONGREN clinical dataset with image sizes of 1880 × 2816. The experimental results show superior performance in comparison to other state-of-theart methods. Especially, for whole-image input from the DRIVE dataset, S-UNet achieved a Matthews correlation coefficient (MCC), an area under curve (AUC), and an F1 score of 0.8055, 0.9821, and 0.8303, respectively. The corresponding scores for the CHASE_DB1 dataset were 0.8065, 0.9867, and 0.8242, respectively. Moreover, our model shows an excellent performance on the TONGREN clinical dataset. In addition, S-UNet segments images of low, medium, and high resolutions in just 33ms, 91ms and 0.49s, respectively. This shows the real-time applicability of the proposed model.
Rolling bearings are widely used in industrial manufacturing, and ensuring their stable and effective fault detection is a core requirement in the manufacturing process. However, it is a great challenge to achieve a highly accurate rolling bearing fault diagnosis because of the severe imbalance and distribution differences in fault data due to weak early fault features and interference from environmental noise. An intelligent fault diagnosis strategy for rolling bearings based on grayscale image transformation, a generative adversative network, and a convolutional neural network was proposed to solve this problem. First, the original vibration signal is converted into a grayscale image. Then more training samples are generated using GANs to solve severe imbalance and distribution differences in fault data. Finally, the rolling bearing condition detection and fault identification are carried out by using SECNN. The availability of the method is substantiated by experiments on datasets with different data imbalance ratios. In addition, the superiority of this diagnosis strategy is verified by comparing it with other mainstream intelligent diagnosis techniques. The experimental result demonstrates that this strategy can reach more than 99.6% recognition accuracy even under substantial environmental noise interference or changing working conditions and has good stability in the presence of a severe imbalance in fault data.
Due to the characteristics of the cotton picker working in the field and the physical characteristics of cotton, it is easy to burn during the operation, and it is difficult to be detected, monitored, and alarmed. In this study, a fire monitoring system of cotton pickers based on GA optimized BP neural network model was designed. By integrating the monitoring data of SHT21 temperature and humidity sensors and CO concentration monitoring sensors, the fire situation was predicted, and an industrial control host computer system was developed to monitor the CO gas concentration in real time and display it on the vehicle terminal. The BP neural network was optimized by using the GA genetic algorithm as the learning algorithm, and the data collected by the gas sensor were processed by the optimized network, which effectively improved the data accuracy of CO concentration during fires. In this system, the CO concentration in the cotton box of the cotton picker was validated, and the measured value of sensor was compared with the actual value, which verified the effectiveness of the optimized BP neural network model with GA. The experimental verification showed that the system monitoring error rate was 3.44%, the accurate early warning rate was over 96.5%, and the false alarm rate and the missed alarm rate were less than 3%. In this study, the fire of cotton pickers can be monitored in real time and an early warning can be made in time, and a new method was provided for accurate monitoring of fire in the field operation of cotton pickers.
The plant factory transplanter is a key component of the plant factory system. Its operation status directly affects the quality and survival rate of planted seedlings, which in turn affects the overall yield and economic efficiency. To monitor the operation status and transplanting quality of a transplanting machine in a timely manner, the primary task is to use a computerized and easy-to-use method to monitor the transplanting units. Inspired by the latest developments in augmented reality and robotics, a digital twin model-based and data-driven online monitoring method for plant factory transplanting equipment is proposed. First, a data-driven and virtual model approach is combined to construct a multi-domain digital twin of the transplanting equipment. Then, taking the vibration frequency domain signal above the transplanting manipulator and the image features of the transplanting seedling tray as input variables, the evaluation method and configuration method of the plant factory transplanter digital twin system are proposed. Finally, the effect of the transplanter is evaluated, and the cycle can be repeated to optimize the transplanter to achieve optimal operation parameters. The results show that the digital twin model can effectively use the sensor data to identify the mechanical vibration characteristics and avoid affecting transplanting quality due to mechanical resonance. At a transplanting rate of 3000 plants/h, the transplanting efficiency can be maintained at a high level and the vibration signal of the X, Y, and Z-axis above the transplanting manipulator is relatively calm. In this case, Combined the optimal threshold method with the traditional Wiener algorithm, the identification rate of healthy potted seedlings can reach 94.3%. Through comprehensively using the optimal threshold method and 3D block matching filtering algorithm for image threshold segmentation and denoising, the recognition rate of healthy seedlings has reached over 96.10%. In addition, the developed digital twin can predict the operational efficiency and optimal timing of the detected transplanter, even if the environmental and sensor data are not included in the training. The proposed digital twin model can be used for damage detection and operational effectiveness assessment of other plant factory equipment structures.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.