《電子技術(shù)應(yīng)用》
您所在的位置:首頁 > 可編程邏輯 > 設(shè)計應(yīng)用 > 一種電纜終端頭紅外識別算法的FPGA實現(xiàn)研究
一種電纜終端頭紅外識別算法的FPGA實現(xiàn)研究
電子技術(shù)應(yīng)用
吳衛(wèi)堃1,鄭耀華1,曾彥超1,曾祥偉1,巫志安1,李嘉成1,周騫2,袁超2
1.廣東電網(wǎng)有限責(zé)任公司肇慶供電局;2.湖南大學(xué) 電氣與信息工程學(xué)院
摘要: 針對在電站巡檢中電纜終端頭識別準(zhǔn)確率低、實時性差等問題,設(shè)計一種基于粒子群算法(Particle Swarm Optimization,PSO)優(yōu)化反向傳播(Back Propagation,BP)神經(jīng)網(wǎng)絡(luò)的現(xiàn)場可編程門陣列(Field Programmable Gate Array,F(xiàn)PGA)紅外識別系統(tǒng)。紅外識別算法實現(xiàn)包括使用改進(jìn)區(qū)域生長算法對紅外圖像進(jìn)行分割,隨后計算Hu不變矩作為神經(jīng)網(wǎng)絡(luò)輸入特征。對于PSO-BP神經(jīng)網(wǎng)絡(luò),選擇7-10-1的網(wǎng)絡(luò)結(jié)構(gòu),訓(xùn)練后均方誤差為0.085,優(yōu)于BP神經(jīng)網(wǎng)絡(luò)的0.136。在FPGA上實現(xiàn)時,采用定點數(shù)據(jù)量化、流水線結(jié)構(gòu)及并行計算方法,同時對Sigmoid激活函數(shù)應(yīng)用二次方程多段擬合。最終經(jīng)過仿真驗證,該系統(tǒng)識別率達(dá)到了92%并且算法速度提高了約6倍。
中圖分類號:TN79 文獻(xiàn)標(biāo)志碼:A DOI: 10.16157/j.issn.0258-7998.246172
中文引用格式: 吳衛(wèi)堃,鄭耀華,曾彥超,等. 一種電纜終端頭紅外識別算法的FPGA實現(xiàn)研究[J]. 電子技術(shù)應(yīng)用,2025,51(7):95-100.
英文引用格式: Wu Weikun,Zheng Yaohua,Zeng Yanchao,et al. Research on FPGA implementation of an infrared identification algorithm for cable terminals[J]. Application of Electronic Technique,2025,51(7):95-100.
Research on FPGA implementation of an infrared identification algorithm for cable terminals
Wu Weikun1,Zheng Yaohua1,Zeng Yanchao1,Zeng Xiangwei1,Wu Zhian1,Li Jiacheng1,Zhou Qian2,Yuan Chao2
1.Zhaoqing Power Supply Bureau, Guangdong Power Grid Co., Ltd.;2.School of Electrical and Information Engineering, Hunan University
Abstract: To address the issues of low identification accuracy and poor real-time performance of cable terminal heads during power station inspections, a field programmable gate array (FPGA) infrared recognition system based on particle swarm optimization (PSO) to optimize back propagation (BP) neural networks has been designed. The infrared recognition algorithm includes the use of an improved region growing algorithm for segmenting infrared images, followed by the calculation of Hu invariant moments as input features for the neural network. For the PSO-BP neural network, a 7-10-1 network structure was chosen, achieving a mean squared error of 0.085 after training, which is better than the 0.136 of the BP neural network. When implemented on the FPGA, fixed-point data quantization, pipelined architecture, and parallel computing methods were employed, along with a piecewise quadratic fitting for the Sigmoid activation function. Ultimately, through simulation verification, the system achieved a recognition rate of 92% and improved the algorithm's speed by approximately six times.
Key words : FPGA;infrared image recognition algorithm;region growing method;Hu invariant moments;PSO-BP neural network

引言

電纜終端頭是電力系統(tǒng)的重要組成部分,其運行狀態(tài)影響電網(wǎng)安全。紅外成像技術(shù)因其非接觸性和穿透力強(qiáng)[1],在電纜終端頭識別中得到廣泛應(yīng)用。然而,復(fù)雜背景使傳統(tǒng)紅外識別算法容易出現(xiàn)識別錯誤、實時性差等問題[2],從而導(dǎo)致其狀態(tài)診斷結(jié)果無法及時匹配對應(yīng)的設(shè)備類型,影響運行狀態(tài)系統(tǒng)的正常運行。

近年來,神經(jīng)網(wǎng)絡(luò)在圖像識別中廣泛應(yīng)用[3],能夠在復(fù)雜環(huán)境中準(zhǔn)確識別目標(biāo)物體。將紅外圖像處理算法與神經(jīng)網(wǎng)絡(luò)結(jié)合是更優(yōu)解。目前,電力巡檢中常用的神經(jīng)網(wǎng)絡(luò)包括BP神經(jīng)網(wǎng)絡(luò)、卷積神經(jīng)網(wǎng)絡(luò)和生成對抗網(wǎng)絡(luò)。然而,這些算法通常依賴計算機(jī)平臺,難以滿足電力巡檢對便攜性和實時性的要求,因此需要移植到嵌入式平臺??紤]到嵌入式硬件資源,BP神經(jīng)網(wǎng)絡(luò)因其簡潔結(jié)構(gòu)和較低計算需求,更適合硬件部署,但其容易陷入局部最優(yōu),而PSO算法可以優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的初始權(quán)重和偏置,顯著提升識別效果[4]。

在硬件設(shè)備的選擇上相比于DSP+ARM架構(gòu),FPGA憑借其并行處理能力和高速計算優(yōu)勢[5],更適合進(jìn)行電站中的數(shù)據(jù)處理[6],近年來已廣泛應(yīng)用于高速圖像處理領(lǐng)域。

本文基于以上分析提出一種基于PSO-BP神經(jīng)網(wǎng)絡(luò)優(yōu)化紅外識別算法的FPGA系統(tǒng),通過FPGA并行計算和流水線結(jié)構(gòu)優(yōu)化,實現(xiàn)電纜終端頭紅外圖像的實時、準(zhǔn)確識別。


本文詳細(xì)內(nèi)容請下載:

http://ihrv.cn/resource/share/2000006602


作者信息:

吳衛(wèi)堃1,鄭耀華1,曾彥超1,曾祥偉1,巫志安1,李嘉成1,周騫2,袁超2

(1.廣東電網(wǎng)有限責(zé)任公司肇慶供電局,廣東 肇慶526000;

2.湖南大學(xué) 電氣與信息工程學(xué)院,湖南 長沙410000)


Magazine.Subscription.jpg

此內(nèi)容為AET網(wǎng)站原創(chuàng),未經(jīng)授權(quán)禁止轉(zhuǎn)載。