Binarized Encoder-Decoder Network and Binarized Deconvolution Engine for Semantic Segmentation

Kim, H; Kim, Jeong-Hoon; Choi, Jungwook; Lee, Jungkeol; Song, Yong Ho

doi:10.1109/access.2020.3048375

Cited by 6 publications

(2 citation statements)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The slightly modified implementation described in the authors' previous work [38] has been used for the AANN implementation. The CU4 and auxiliary arithmetic blocks perform operations described by ( 13) -( 21), as well as they calculate the novelty assessment level coefficient (22). The comparison of the coefficient above with the threshold value δ (11) is, in turn, accomplished by the CU1 main control unit.…”

Section: Architecture and Hardware Implementationmentioning

confidence: 99%

“…The proposed method is more power-efficient as compared to the general-purpose processors performing the same algorithm. A binarized encoderdecoder network and binarized deconvolution engine to enable low-power, real-time semantic segmentation was proposed in [22]. The binarized deconvolution engine was implemented on FPGA and accelerated the proposed encoderdecoder within CamVid 11 images achieving notable performance.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

FPGA-Embedded Anomaly Detection System for Milling Process

et al. 2021

View full text Add to dashboard Cite

The main goal of this work is to design a supervising controller able to detect an anomaly in the milling process and implement the soultion in Field Programmable Gate Array (FPGA) chip. Executing this task, the controller continuously monitors the vibration signal coming from the acceleration sensor, installed on the milling machine, and striving to isolate new vibration patterns which are different from typical patterns recorded for the correct milling process. The detection method relies on determining selected signal features in the frequency domain and applying an auto-associative neural network (AANN) for novelty detection. It has been shown that by exercising the frequency spectrum of the vibration signal, extracting specific features of the signal's spectrum, and using an auto-associative neural network, it is possible to detect anomalies in a milling process with relatively high efficiency. The accuracy, sensitivity, specificity, precision, and false alarm rate are equal to 94.3, 100, 91.2, 88.9, and 8.8 percent. All necessary calculations can be accomplished by the developed single-chip FPGA embedded supervising controller. The controller allows high-speed calculations under low power consumption. It characterizes high reliability and low price compared to typically encountered solutions.

show abstract

Section: Architecture and Hardware Implementationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%