Shipping containers play an important role in global transportation. As container codes are the unique identifiers for shipping containers, recognizing these codes is an essential step to manage the containers and logistics. The conventional code localization methods can easily be interfered by varied noises and cannot identify the best regions for code recognition. In this article, we propose an adaptive deep learning framework for shipping container code localization and recognition. In the framework, the noisy text regions will be removed by an adaptive score aggregation (ASA) algorithm. The code region boundaries are identified by the average-to-maximum suppression range (AMSR) algorithm. Thus, the predicted locations can be adjusted within this range to fit the code recognition model to achieve higher accuracy. The experimental results on the comparative study with the state-of-the-art models, including EAST, PSENet, GCRNN, and MaskTextSpotter, demonstrated that the proposed framework achieved better localization performance and obtained 93.33% recognition accuracy. The processing speed reaches 1.13 frames/s, which is sufficient to meet the operational requirements. Thus, the proposed solution will facilitate the digital transformation of shipping container management and logistics at ports.

An Adaptive Deep Learning Framework for Shipping Container Code Localization and Recognition

New paradigm of FPGA-based computational intelligence from surveying the implementation of DNN accelerators

To reduce the extra work, the operation cost, and the risk of cargo delay induced by the unloading of wrong containers, this study first develops a container color detection model to predict the color of the container being unloaded. The prediction results are then used to develop two crane operator alarm methods. Method 1 alerts the crane operator if the detected color of a container is not in compliance with the correct container color. Method 2 constructs a decision problem to decide whether to alert the operator. The results of numerical experiments show that methods 1 and 2 are better than the benchmark. Specifically, method 1 can save the expected annual total cost by about 82% while method 2 can save the expected annual total cost by about 85%. Extensive sensitivity analysis is also conducted to verify the methods performance and robustness. 

Development of computer vision informed container crane operator alarm methods

The shipping industry is a multifaceted trading system, enabling the flow of goods between countries. Often the standards, methods and technologies become interspersed and difficult to plan, disrupting the supply chain. Data is not always accurate, necessitating further analysis. This paper presents an approach to solving key issues in two important facets of the supply chain, predicting the date of restitution for a cargo container and using an optical character recognition (OCR)-centred pipeline to extrapolate data from containers. Both approaches use long short-term memory (LSTM) models, including bi-directional LSTMs. These methods leverage state-of-the-art text recognition architecture and advanced algorithm ensembling, giving significant improvements in data quality. The experimental results illustrate that these methods vastly outperform current industry practices and more recent approaches.

Improving Data Quality in the Cargo Industry with Modern Recurrent Neural Network Architecture

Automatic container-code recognition system becomes a crucial requirement for ship transportation industry in recent years. In this paper, an automatic container-code recognition system based on computer vision and deep neural networks is proposed. The system consists of two modules, detection module and recognition module. The detection module applies both algorithms based on computer vision and neural networks, and generates a better detection result through combination to avoid the drawbacks of the two methods. The combined detection results are also collected for online training of the neural networks. The recognition module exploits both character segmentation and end-to-end recognition, and outputs the recognition result which passes the verification. When the recognition module generates false recognition, the result will be corrected and collected for online training of the end-to-end recognition sub-module. By combining several algorithms, the system is able to deal with more situations, and the online training mechanism can improve the performance of the neural networks at runtime. The proposed system is able to achieve 93% of overall recognition accuracy.

Container-code recognition system based on computer vision and deep neural networks

Most FPGA accelerators for convolutional neural network are designed to optimize the inner acceleration and are ignored of the optimization for the data path between the inner accelerator and the outer system. This could lead to poor performance in applications like real time video object detection. We propose a brand new systematic FPFA acceleration design to solve this problem. This design takes the data path optimization between the inner accelerator and the outer system into consideration and optimizes the data path using techniques like hardware format transformation, frame compression. It also takes fixed-point, new pipeline technique to optimize the inner accelerator. All these make the final system’s performance very good, reaching about 10 times the performance comparing with the original system.Most FPGA accelerators for convolutional neural network are designed to optimize the inner acceleration and are ignored of the optimization for the data path between the inner accelerator and the outer system. This could lead to poor performance in applications like real time video object detection. We propose a brand new systematic FPFA acceleration design to solve this problem. This design takes the data path optimization between the inner accelerator and the outer system into consideration and optimizes the data path using techniques like hardware format transformation, frame compression. It also takes fixed-point, new pipeline technique to optimize the inner accelerator. All these make the final system’s performance very good, reaching about 10 times the performance comparing with the original system.

A systematic FPGA acceleration design for applications based on convolutional neural networks

The present invention relates to a carbon nanometer transistor memory test method. The method comprises a first step of testing a carbon nanometer transistor memory by using a leap testing algorithm, and rapidly locating an error segment, so as to obtain error distribution information of the memory; a second step of performing redundancy analysis on the obtained error information by using a redundancy analysis structure, so as to obtain a repairable chip and an irreparable chip; and a third step of performing a marching test on the repairable chip and untested positions so as to test all storage units to obtain accurate error information. CNFET-SRAM chips having serious cascading faults are screened out by using a test method provided by the present invention in an early age, so that the marching test cost in a later period is saved. The quantity of tested storage units is effectively reduced, the test method is simple to implement, the components are analyzed through cooperation with redundant resources, whether the chip can be repaired is determined rapidly, the detection speed is improved, and unnecessary overhead in the later period is reduced.

Novel carbon nanometer transistor memory test method

To deal with the considerable deviation of transparency tracing method and digital planimetry method used in current clinical diabetic foot ulcer injury assessment, this paper proposes a 3D reconstruction system which can be used to get foot model with good quality texture, then injury assessment is done by measuring the reconstructed model. The system uses the Intel RealSense SR300 depth camera which is based on infrared structured-light as input device, the required data from different view is collected by moving the camera around the scanned object. The geometry model is reconstructed by fusing the collected data, then the mesh is sub-divided to increase the number of mesh vertices and the color of each vertex is determined using a non-linear optimization, all colored vertices compose the surface texture of the reconstructed model. Experimental results indicate that the reconstructed model has millimeter-level geometric accuracy and texture with few artificial effect.

/pdf/application-of-3d-reconstruction-system-in-diabetic-foot-4ii5z5gu08.pdf

Xiaoyao Liang

Papers

Container-code recognition system based on computer vision and deep neural networks

A systematic FPGA acceleration design for applications based on convolutional neural networks

Novel carbon nanometer transistor memory test method

Application of 3D reconstruction system in diabetic foot ulcer injury assessment