何靜飛 張瀟月 周亞同
摘要 隨物聯(lián)網(wǎng)戰(zhàn)略地位和影響力的不斷提升,無線傳感器網(wǎng)絡(luò)(Wireless Sensor Networks, WSNs)作為物聯(lián)網(wǎng)核心技術(shù)之一,迎來了一場新的研究熱潮。如何降低網(wǎng)絡(luò)能耗,延長網(wǎng)絡(luò)生命周期一直是WSNs研究的關(guān)鍵問題。近年來,隨壓縮感知及低秩理論的提出,基于數(shù)據(jù)稀疏表示的WSNs數(shù)據(jù)匯聚方法受到廣泛關(guān)注。利用WSNs數(shù)據(jù)的高度時空冗余特性,可有效降低數(shù)據(jù)傳輸量,降低網(wǎng)絡(luò)能耗。本文從幾個方面介紹現(xiàn)有基于數(shù)據(jù)稀疏表示的WSNs數(shù)據(jù)匯聚方法:首先,介紹壓縮感知理論模型及壓縮數(shù)據(jù)匯聚框架,分別從稠密隨機投影和稀疏隨機投影角度介紹基于壓縮感知的數(shù)據(jù)匯聚方法;然后,介紹矩陣補全理論模型和基于矩陣補全的數(shù)據(jù)匯聚及重建方法;最后,提出無線傳感器網(wǎng)絡(luò)數(shù)據(jù)匯聚存在的問題和對未來研究趨勢的展望。
關(guān) 鍵 詞 無線傳感器網(wǎng)絡(luò);稀疏表示;壓縮感知;矩陣補全;數(shù)據(jù)收集;時空相關(guān)性
中圖分類號 TP212.9;TN929.5? ? ?文獻標志碼 A
Abstract With continuous improving of the strategic position and influence of Internet of Things, Wireless Sensor Networks (WSNs), as one of the core technologies of Internet of Things, has become a research focus. So how to reduce the network energy consumption and prolong the network lifetime has been a key research issue in WSNs. Recently, with the development of compressed sensing and low rank theory, the data aggregation method based on sparse representation in WSNs has attracted much attention. By exploiting the high spatiotemporal redundancy of WSNs data, the amount of data transmitted is effectively reduced and network energy consumption is reduced. This paper introduces the existing data aggregation methods of WSNs based on sparse representation. First, compressed sensing model and compressed data collection framework are introduced. Specifically, data aggregation methods based on compressed sensing are introduced from the perspectives of dense random projection and sparse random projection. Then matrix completion model and data aggregation and reconstruction methods based on matrix completion are discussed. Finally, the existing problems of data aggregation in WSNs and the prospect of future research are mentioned.
Key words wireless sensor network; sparse representation; compressed sensing; matrix completion; data gathering; spatio-temporal correlation
0 引言
能量消耗是無線傳感器網(wǎng)絡(luò)(Wireless Sensor Networks, WSNs)[1]中最為重要的問題。隨著壓縮感知和低秩理論的提出,基于稀疏表示的WSNs數(shù)據(jù)匯聚方法受到科研人員的廣泛關(guān)注。通過利用WSNs數(shù)據(jù)的高度時空冗余性來有效降低數(shù)據(jù)傳輸量,基于稀疏表示的WSNs數(shù)據(jù)匯聚方法取得了巨大的成果。鑒于此,本文系統(tǒng)的對基于稀疏表示的WSNs數(shù)據(jù)匯聚方法進行分類介紹。
1 介紹
伴隨著信息時代的到來,物聯(lián)網(wǎng)(Internet of Things, IoT)[2]這一新興信息產(chǎn)業(yè)迎來了發(fā)展的熱潮,并逐漸改變著人們的生活方式。無線傳感器網(wǎng)絡(luò)是物聯(lián)網(wǎng)的核心技術(shù)之一,因物聯(lián)網(wǎng)的迅速發(fā)展而受到研究人員廣泛的關(guān)注,開啟了無線傳感器網(wǎng)絡(luò)發(fā)展的新紀元。
WSNs由基站和大量傳感器節(jié)點構(gòu)成,采用自組織方式連接。如圖1所示,傳感器節(jié)點部署在探測區(qū)域內(nèi),采用多跳方式將感知數(shù)據(jù)發(fā)送至基站,基站將數(shù)據(jù)處理后提供給用戶使用。WSNs因具有獨立的數(shù)據(jù)采集、處理和傳輸能力,被廣泛應(yīng)用在各個領(lǐng)域,包括軍事國防[3]、環(huán)境監(jiān)測[4]、健康醫(yī)療[5]和工農(nóng)業(yè)[6]等。隨著社會生產(chǎn)和日常生活對WSNs需求的增加,如何大規(guī)模高密度的部署WSNs成為研究熱點。然而WSNs中網(wǎng)絡(luò)能耗不均勻?qū)е虏糠止?jié)點過早死亡,節(jié)點容量、數(shù)據(jù)處理和存儲能力有限等問題,阻礙著WSNs的大規(guī)模高密度部署。近年來,隨信號處理領(lǐng)域中壓縮感知(Compressed Sensing, CS)[7-8]及低秩理論[9-10]的提出與發(fā)展,基于稀疏表示的WSNs數(shù)據(jù)匯聚方法因采用的信號處理方式能夠有效降低網(wǎng)絡(luò)能耗而受到廣泛關(guān)注。根據(jù)現(xiàn)有研究,基于稀疏表示的WSNs數(shù)據(jù)匯聚方法可分為兩大類:1)基于約束數(shù)據(jù)一階稀疏性,即壓縮感知;2)基于約束數(shù)據(jù)二階稀疏性,即低秩約束。
在實際應(yīng)用部署中,WSNs中大量傳感器節(jié)點密集分布在一定區(qū)域內(nèi),節(jié)點在一段連續(xù)時間內(nèi)感知的數(shù)據(jù)往往具有很高的時空冗余度,在特定稀疏基下具有稀疏性,符合CS的應(yīng)用前提。因此,CS的提出開啟了WSNs數(shù)據(jù)匯聚方法的新篇章。基于CS的WSNs數(shù)據(jù)匯聚方法允許網(wǎng)絡(luò)以低于奈奎斯特頻率的方式采集數(shù)據(jù),將采集與壓縮同時進行,通過減少WSNs數(shù)據(jù)傳輸量降低網(wǎng)絡(luò)能耗。基于CS的WSNs數(shù)據(jù)匯聚方法受到了廣泛關(guān)注并取得了巨大成功。受CS理論的推動,本質(zhì)為約束數(shù)據(jù)二階稀疏性的低秩矩陣模型[9, 10]隨之興起,并由此系統(tǒng)地發(fā)展出了新的理論與應(yīng)用。而后,科研人員將低秩矩陣應(yīng)用到WSNs中,基于低秩矩陣的數(shù)據(jù)匯聚方法是將WSNs中不同節(jié)點在不同時刻采集的數(shù)據(jù)分布到一個環(huán)境矩陣中,通過稀疏采樣方式僅采集部分數(shù)據(jù),由部分數(shù)據(jù)重建出原始數(shù)據(jù)。該過程在數(shù)學(xué)形式上是一個矩陣補全(Matrix Completion, MC)問題。由于數(shù)據(jù)本身具有時空相關(guān)性,因此基站可通過約束矩陣低秩性恢復(fù)原始數(shù)據(jù)矩陣。相比較于CS,基于MC的數(shù)據(jù)匯聚方法更有效地利用了WSNs數(shù)據(jù)的時空相關(guān)性,能夠獲得更高的數(shù)據(jù)重建精度,因此更加受到科研人員的關(guān)注。
本文將系統(tǒng)地介紹現(xiàn)有基于數(shù)據(jù)稀疏表示的WSNs數(shù)據(jù)匯聚方法,第2節(jié)將介紹壓縮感知理論及基于壓縮感知的WSNs數(shù)據(jù)匯聚方法,第3節(jié)介紹矩陣補全及基于矩陣補全的數(shù)據(jù)匯聚和重建方法,第4節(jié)對基于數(shù)據(jù)稀疏表示的WSNs數(shù)據(jù)匯聚方法的研究前景進行展望,第5節(jié)進行總結(jié)。
2 基于壓縮感知的數(shù)據(jù)匯聚
2.1 壓縮感知基本理論
壓縮感知理論于2006年由Donoho[8], Cands[7]等提出,其核心思想是將采樣和壓縮同時進行。在壓縮感知框架中,原始信號[x∈?n×1]為稀疏信號或可稀疏信號,通過測量矩陣[Φ∈?m×n]可獲得測量值[y=Φx∈?m×1],這里[m?n]。由測量值[y]即可通過重構(gòu)算法重建出原始信號[x],即求解如式(1)的[?0]最小化問題:
式中,[ψ∈?n×n]為稀疏基矩陣,CS理論通過約束原始信號[x]在一定稀疏基[ψ]下的稀疏性,實現(xiàn)欠定問題的求解。然而[?0]范數(shù)問題的求解是一個NP-hard(Nondeterministic Polynomial-time Hard)問題[11],由于[?1]范數(shù)是[?0]范數(shù)的凸包絡(luò),式(1)可轉(zhuǎn)化為約束[?1]范數(shù)最小化[12]:
式(2)為凸優(yōu)化問題,因此數(shù)據(jù)的重構(gòu)過程轉(zhuǎn)換為求解一線性優(yōu)化問題。CS中的數(shù)據(jù)重構(gòu)算法包括凸松弛類算法[13]、貪婪類算法[14-15]和迭代閾值類算法[16]。
2.2 壓縮數(shù)據(jù)收集框架的建立
實際中,WSNs采集到的數(shù)據(jù)因節(jié)點部署密集具有較高的時空相關(guān)性,在一定稀疏基下呈現(xiàn)稀疏性,為可稀疏信號。鑒于此,科研學(xué)者嘗試將CS理論應(yīng)用于WSNs數(shù)據(jù)匯聚中,并取得了一定成功。而后,眾多基于CS的WSNs數(shù)據(jù)匯聚方法應(yīng)運而生[17-20]。
受到無線通信和CS理論研究啟發(fā),文獻[21]首次將CS理論應(yīng)用到WSNs中,提出壓縮無線傳感的概念(Compressive Wireless Sensing, CWS),用于估計具有某種結(jié)構(gòu)規(guī)律的傳感器數(shù)據(jù)能量有效性。文獻[22]首次將CS理論應(yīng)用到大規(guī)模WSNs數(shù)據(jù)匯聚中,提出了壓縮數(shù)據(jù)采集模型(Compressive Data Gathering, CDG),為WSNs中的數(shù)據(jù)匯聚方法開辟了一條新道路。
圖2a)為無壓縮數(shù)據(jù)匯聚方法中通過多跳方式將節(jié)點數(shù)據(jù)轉(zhuǎn)發(fā)到基站的直觀圖,節(jié)點[N1]將其感知數(shù)據(jù)[x1]發(fā)送到[N2],[N2]將其感知數(shù)據(jù)[x2]和接收到的[x1]發(fā)送到[N3],以此類推。在路由結(jié)束時,[Nn]將[n]個數(shù)據(jù)發(fā)送到基站??梢?,距離基站越近的節(jié)點消耗能量越多,網(wǎng)絡(luò)存在能耗不均勻情況,整體數(shù)據(jù)傳輸量為[n(n+1)2]。圖2b)展示了CDG數(shù)據(jù)傳輸方式,節(jié)點[N1]產(chǎn)生一個隨機向量[φ1=(?11,?21,…,?m1)∈?m×1],將其感知數(shù)據(jù)[x1]乘以[φ1]后發(fā)送到[N2]。[N2]接收到數(shù)據(jù)后,將其自身感知數(shù)據(jù)[x2]乘以隨機向量[φ2]后將[x1φ1+x2φ2]發(fā)送到[N3],以此類推。第[i]個節(jié)點[Ni]所需傳輸數(shù)據(jù)為[j=1ixjφj],網(wǎng)絡(luò)整體數(shù)據(jù)傳輸量為[mn]。相較于無壓縮采集數(shù)據(jù)方法,CDG中每個節(jié)點傳輸?shù)臄?shù)據(jù)是相同長度的,能夠均衡網(wǎng)絡(luò)中的能量消耗,避免因少數(shù)節(jié)點過早死亡減少網(wǎng)絡(luò)的生命周期,同時降低了網(wǎng)絡(luò)整體數(shù)據(jù)傳輸量。圖3a)展示了CDG使用樹形拓撲結(jié)構(gòu)傳輸數(shù)據(jù)的路由圖,其中假設(shè)測量值維數(shù)為5。
而后,文獻[23]對CDG進行改進,提出IR-CDG傳輸機制,采用[Φ=I R∈?M×N]形式的測量矩陣,其中[I]為單位矩陣,[R∈?M×(N-M)]為高斯隨機矩陣,[Φ]結(jié)構(gòu)如式(3):
該測量矩陣具有良好的約束等距性原則(Restricted Isometry Property, RIP),在保證數(shù)據(jù)可靠性的情況下,可通過減少參與傳輸節(jié)點的方式來降低通信成本,但單個測量值的收集需要傳輸?shù)臄?shù)據(jù)量仍然與CDG相同。文獻[24]再次對CDG做出改進,提出一種混合壓縮感知(Hybrid-CS)數(shù)據(jù)采集方法。如圖3b)所示,Hybrid-CS中只有當轉(zhuǎn)發(fā)量大于自身測量值的維數(shù)時才進行加權(quán)處理,否則直接轉(zhuǎn)發(fā)自身測量值到父節(jié)點即可,以此減少單個測量值收集時需要傳輸?shù)臄?shù)據(jù)量。
隨著研究的逐漸深入,越來越多的基于CS的數(shù)據(jù)匯聚方法被提出。這些方法根據(jù)測量矩陣類型可分為兩大類:1)基于稠密投影的WSNs數(shù)據(jù)匯聚方法,前面所述的CDG即為該類代表性方法,其特點在于測量矩陣中元素值幾乎全不為零值;2)基于稀疏投影的WSNs數(shù)據(jù)匯聚方法,IR-CDG可認為是該類方法的雛形,其特點在于測量矩陣中元素值存在一定數(shù)量的零值。如圖3c)所示,在一次數(shù)據(jù)采集時隙中,只有5個節(jié)點進行數(shù)據(jù)采集,其他節(jié)點處于休眠狀態(tài),以此來降低網(wǎng)絡(luò)能耗。后續(xù)將分別介紹基于稠密投影和基于稀疏投影的WSNs數(shù)據(jù)匯聚方法。
2.3 基于稠密投影的WSNs數(shù)據(jù)匯聚方法
通過上述的CDG及其改進算法可看出,CDG框架是通過減少數(shù)據(jù)傳輸量來降低能耗,延長網(wǎng)絡(luò)壽命。而后,多種基于稠密隨機投影的數(shù)據(jù)匯聚方法相繼提出。
其中,部分文獻從路由優(yōu)化方面對數(shù)據(jù)匯聚算法進行改進[25-28]。針對同構(gòu)網(wǎng)絡(luò),文獻[25]提出了一種適用于CS的分布式數(shù)據(jù)匯聚方案,該方案中每個傳感器節(jié)點獨立地找到其父節(jié)點并構(gòu)造路由樹的一部分,而不需要中心單元來構(gòu)造所有轉(zhuǎn)發(fā)樹。考慮到能量消耗對數(shù)據(jù)匯聚的影響,文獻[26]提出基于能量感知的CS數(shù)據(jù)匯聚和能量平衡的高層數(shù)據(jù)聚合樹兩種數(shù)據(jù)匯聚方法,在平衡節(jié)點間能量消耗的同時提高了網(wǎng)絡(luò)的生存周期。為解決傳統(tǒng)的基于CS的數(shù)據(jù)匯聚方案不適用于異構(gòu)WSNs的問題,文獻[27]提出一種基于CS的聚類、非均勻分層、多跳路由算法。采用改進的LEACH協(xié)議實現(xiàn)簇間傳輸和融合,然后采用非均勻分層多跳路由將融合后的數(shù)據(jù)包轉(zhuǎn)發(fā)到基站進行數(shù)據(jù)重建。
為提高CS數(shù)據(jù)匯聚算法中的重建精度,眾多關(guān)于重建算法的研究文獻相繼發(fā)表。文獻[29]提出雙層壓縮聚合(Dual-lEvel Compressed Aggregation, DECA)技術(shù)。DECA分為兩個層級重建數(shù)據(jù),在第一級使用基于擴散小波的CS來重建數(shù)據(jù)。在第二級,DECA使用矩陣補全對第一級的結(jié)果數(shù)據(jù)再次進行重建。文獻[30]提出一種基于聯(lián)合稀疏性的壓縮感知技術(shù),采用貝葉斯推理來建立信號的概率模型。在該貝葉斯推理恢復(fù)框架中,使用置信傳播算法作為解碼方法來壓縮和重建空間相關(guān)信號。
此外,部分研究通過利用WSNs信號中的其他特性,進一步提高數(shù)據(jù)的重建精度。文獻[31]將自回歸模型引入到感知數(shù)據(jù)重建中,利用感知數(shù)據(jù)的局部相關(guān)性,實現(xiàn)了感知數(shù)據(jù)的局部自適應(yīng)稀疏性。通過調(diào)整目標函數(shù)中的自回歸參數(shù)使得數(shù)據(jù)重建適應(yīng)于感知到的數(shù)據(jù),并且根據(jù)數(shù)據(jù)的變化自適應(yīng)的調(diào)節(jié)所感測數(shù)據(jù)需要的測量次數(shù)。文獻[32]將全變分(Total Variation,TV)模型引入到CDG中,利用WSNs數(shù)據(jù)在不同方向上梯度的稀疏性,提出一種壓縮的多時隙數(shù)據(jù)采集方法來采集任意數(shù)量的數(shù)據(jù)。然后利用TV正則化方法構(gòu)造數(shù)據(jù)重建算法,并提出一種交替最小化方法來求解重建算法。
2.4 基于稀疏投影的WSNs數(shù)據(jù)匯聚方法
基于CS的數(shù)據(jù)匯聚方法中的數(shù)據(jù)傳輸量取決于測量矩陣,相對于稠密隨機矩陣,稀疏隨機投影矩陣中大部分元素值為零值,零值元素對應(yīng)位置的節(jié)點數(shù)據(jù)無需轉(zhuǎn)發(fā),因此數(shù)據(jù)傳輸能耗大大降低。
文獻[19]針對樹型拓撲網(wǎng)絡(luò),提出最稀疏測量矩陣,該矩陣中每一行只有一個非零項,將單個測量值所需的節(jié)點數(shù)量降到最低。針對集群型網(wǎng)絡(luò),多種數(shù)據(jù)匯聚方法相繼提出[33-35],文獻[33]提出一種最稀疏數(shù)據(jù)采集方案,該方案中只有節(jié)點的一個隨機子集收集數(shù)據(jù),每個節(jié)點的感知數(shù)據(jù)被視為一次測量值。文獻[34]提出一種空間相關(guān)的基站輔助集群,在該集群中僅簇頭感知數(shù)據(jù),然后將數(shù)據(jù)傳輸?shù)交?,不產(chǎn)生簇內(nèi)的通信開銷。然而,網(wǎng)絡(luò)中常因鏈路不可靠導(dǎo)致數(shù)據(jù)丟失,現(xiàn)有方法通常直接將測量矩陣與網(wǎng)絡(luò)路由被動匹配,這樣會使得參與單個測量值的節(jié)點過多,測量值易受網(wǎng)絡(luò)丟包影響。為減少網(wǎng)絡(luò)丟包對測量值的破壞,文獻[36]提出一種聯(lián)合路由的稀疏隨機投影數(shù)據(jù)匯聚方案,該方案先是依據(jù)路由設(shè)計稀疏投影矩陣來減少丟失數(shù)據(jù)對測量值的損毀,然后設(shè)計稀疏投影矩陣約束下的低相干稀疏表示基來確保數(shù)據(jù)的重建精度。
為達到數(shù)據(jù)匯聚算法自適應(yīng)于網(wǎng)絡(luò),根據(jù)網(wǎng)絡(luò)的變化改變參數(shù)來提高采樣質(zhì)量的目的,許多自適應(yīng)算法相繼提出[20,37-39]。為自適應(yīng)調(diào)整取樣率,文獻[37]提出一種基于自適應(yīng)CS的WSNs樣本調(diào)度機制,該機制在每個采樣窗口的基礎(chǔ)上,根據(jù)給定的傳感質(zhì)量估計所需的最小采樣率,并相應(yīng)地調(diào)整傳感器的采樣率。文獻[39]提出一種自適應(yīng)壓縮傳感數(shù)據(jù)匯聚方案,只需對少量測量數(shù)據(jù)重新采樣,即可確定當前稀疏度和新的采樣率。為提高重建算法的恢復(fù)性能和收斂速度,提出了一種結(jié)合稀疏自適應(yīng)匹配追蹤算法的自適應(yīng)步長變化算法。
考慮到使用CS收集數(shù)據(jù)時,WSNs的拓撲結(jié)構(gòu)和路由機制也是需要關(guān)注的重點問題,多篇相關(guān)研究相繼提出。文獻[40]提出最小生成樹的投影(Minimum Spanning Tree Projection, MSTP),該方法中以隨機選擇的投影節(jié)點為根,組建一個最小生成樹(Minimum-Spanning-Trees, MSTs),每個投影節(jié)點將接收的數(shù)據(jù)的加權(quán)和發(fā)送到基站。針對集群式WSNs,文獻[41]利用隨機采樣和隨機游走相結(jié)合的方式構(gòu)造稀疏的二值感知矩陣。CS用來收集WSNs中空間域的感知數(shù)據(jù),隨機抽樣和隨機游走分別用來選擇集群和簇頭中的感知數(shù)據(jù)。而后,文獻[42]中進一步改進,隨機抽樣和隨機游走分別在時間域和空間域采集數(shù)據(jù)。文獻[43]提出一種同時利用時空相關(guān)性的結(jié)合Kronecker壓縮感知和簇拓撲結(jié)構(gòu)的數(shù)據(jù)匯聚方法,簇頭基于收集的數(shù)據(jù)生成稀疏的自測量矩陣,基站將這些矩陣構(gòu)造為塊對角矩陣,以利用簇間的空間相關(guān)性。
相較于稠密投影,基于稀疏投影的WSNs數(shù)據(jù)匯聚方法更加節(jié)能網(wǎng)絡(luò)能耗,因此受到研究人員的青睞。
3 低秩理論在WSNs數(shù)據(jù)匯聚中的研究
3.1 矩陣補全基本理論
受壓縮感知理論取得的巨大成功的推動,本質(zhì)為約束數(shù)據(jù)二階稀疏性的低秩矩陣模型[9,44]近年來獲得研究人員的關(guān)注。低秩矩陣模型中一個典型應(yīng)用即為矩陣補全,當矩陣[M∈?m×n]中僅有部分觀測數(shù)據(jù)已知,而其他數(shù)據(jù)均缺少時,利用部分觀測數(shù)據(jù)恢復(fù)出矩陣[M]的過程即為矩陣補全。如同壓縮感知的前提是數(shù)據(jù)稀疏性,矩陣補全的前提是矩陣低秩性。通過約束矩陣的秩,恢復(fù)出原始矩陣:
式中:[X]為決策矩陣;[rankX]表示矩陣[X]的秩;[Π]代表矩陣中觀測到的數(shù)據(jù)所在位置索引的集合。類似于向量中[?0]最小化問題,矩陣秩最小化問題也是一個NP-hard問題。上述壓縮感知中采用[?1]范數(shù)替代[?0]范數(shù),這里,可采用核范數(shù)替代秩最小化問題[45],式(5)可轉(zhuǎn)化為
式中[·*]為核范數(shù),即矩陣的奇異值之和。低秩矩陣模型目前已廣泛應(yīng)用在信號處理和機器學(xué)習等領(lǐng)域。由于WSNs中不同節(jié)點在不同采集時刻獲取的數(shù)據(jù)可構(gòu)成矩陣形式,數(shù)據(jù)的時空相關(guān)性使得矩陣呈現(xiàn)低秩特性,為矩陣補全應(yīng)用到WSNs數(shù)據(jù)匯聚提供了前提條件。
3.2 WSNs數(shù)據(jù)的時空相關(guān)性
WSNs環(huán)境感知由密集部署在一定區(qū)域的節(jié)點完成,可監(jiān)測風力、溫度、濕度、光照等環(huán)境信息。假設(shè)一個WSN由一個基站和[n]個傳感器節(jié)點組成,用[N1,N2,…,Nn]表示節(jié)點序號,節(jié)點每隔時間[τ]感知一次數(shù)據(jù)并傳輸?shù)交?,因此在[mτ]的時間范圍內(nèi),由[n]個節(jié)點采集到的[m×n]個數(shù)據(jù)可組成環(huán)境矩陣[M]:
如圖4所示,[M]的行由同一節(jié)點在不同采集時刻的感知數(shù)據(jù)組成,列由不同節(jié)點在同一時刻的感知數(shù)據(jù)組成。由于節(jié)點分布密集,數(shù)據(jù)采集時刻間隔短,鄰近節(jié)點在相鄰時刻感知到的數(shù)據(jù)往往波動很小,即這些數(shù)據(jù)通常具有較高的時空相關(guān)性。也就是說[M]具有低秩特性,符合矩陣補全應(yīng)用的前提。因此可通過稀疏采樣收集數(shù)據(jù),然后基于矩陣補全對缺失數(shù)據(jù)進行重建,以此來降低網(wǎng)絡(luò)能耗。
3.3 基于矩陣補全的WSNs數(shù)據(jù)匯聚方法
因WSNs數(shù)據(jù)具有低秩性,滿足矩陣補全理論應(yīng)用的前提,近年來研究人員將MC成功應(yīng)用于WSNs的數(shù)據(jù)匯聚與重建中[46-56]。
2010年,Cheng等[48]首次將MC應(yīng)用到WSNs數(shù)據(jù)匯聚中,提出Efficient Data Gathering Approach (EDCA)。為減少網(wǎng)絡(luò)能耗,隨機選擇部分節(jié)點進行數(shù)據(jù)感知,而后將數(shù)據(jù)直接發(fā)送到基站。數(shù)據(jù)重建上,EDCA將MC中秩最小化問題轉(zhuǎn)換為凸優(yōu)化問題。通過稀疏采樣方式減少了網(wǎng)絡(luò)數(shù)據(jù)傳輸量,在一定程度上降低了網(wǎng)絡(luò)能耗。然而當采樣率極低時,稀疏采樣構(gòu)成的矩陣存在空列的概率極高,此時采用EDCA會導(dǎo)致較大的數(shù)據(jù)重建誤差。為解決這個問題,文獻[49]提出Spatio-Temporal Compressive Data Collection (STCDG)方法,同時利用矩陣的低秩性和數(shù)據(jù)短時間內(nèi)穩(wěn)定性進一步降低數(shù)據(jù)傳輸量。針對降低采樣率導(dǎo)致的空列問題,STDCG在重構(gòu)矩陣時會先刪除空列,只恢復(fù)非空列,然后使用基于時間穩(wěn)定性的優(yōu)化技術(shù)恢復(fù)空列,以此來提高數(shù)據(jù)的重建精度。在進一步利用信號時空相關(guān)性的全局和局部相關(guān)特性基礎(chǔ)上,文獻[57]提出一種基于低階差分平滑度的數(shù)據(jù)重建算法,將時變圖形信號的差分平滑先驗引入到時空信號分析領(lǐng)域。
而后,為進一步提高數(shù)據(jù)重建精度,降低網(wǎng)絡(luò)數(shù)據(jù)傳輸量,多篇相關(guān)研究相繼提出。文獻[50]通過對WSNs數(shù)據(jù)稀疏性和低秩性的研究,提出一種聯(lián)合稀疏約束和低秩約束的無線傳感器網(wǎng)絡(luò)數(shù)據(jù)匯聚重建模型,并提出基于加性半二次正則化方法的重構(gòu)算法,保證了數(shù)據(jù)重建的運算速度與準確性。文獻[58]提出一種負載均衡的隨機采樣機制,通過變動的采樣率在確保重建精度的前提下降低網(wǎng)絡(luò)能耗,同時利用快速奇異值閾值算法提高了重建速度。
在WSNs數(shù)據(jù)匯聚方法上,文獻[51]通過引入歷史數(shù)據(jù),將MC與CDG算法相結(jié)合,在CDG數(shù)據(jù)匯聚框架下,約束歷史數(shù)據(jù)與當前時刻數(shù)據(jù)構(gòu)成矩陣的低秩性,在CDG數(shù)據(jù)匯聚框架下大幅度提高了數(shù)據(jù)重建精度。文獻[52]提出一種基于MC的實時數(shù)據(jù)匯聚方法(MC-Weather)。該方法包括一種基于三種樣本學(xué)習原則的自適應(yīng)采樣算法,能夠快速找到一個適用于矩陣補全的稀疏采樣數(shù)據(jù)集。MC-Weather在動態(tài)環(huán)境中以較低的通信成本成功實現(xiàn)了更高的數(shù)據(jù)重建精度。文獻[59]提出一種適用于集群型網(wǎng)絡(luò)的實時稀疏數(shù)據(jù)匯聚重建方法,所提出的稀疏匯聚方案實現(xiàn)了存在部分節(jié)點休眠情況下的數(shù)據(jù)匯聚,通過先前重建數(shù)據(jù)估計出代表空間分布的子空間,結(jié)合全變分約束實現(xiàn)了實時WSNs數(shù)據(jù)的高精度重建。
同時,鑒于矩陣補全的特征,科研學(xué)者也將其應(yīng)用到WSNs數(shù)據(jù)丟失和損壞中。在WSNs中,受硬件和無線條件的影響,原始感知數(shù)據(jù)通常會出現(xiàn)明顯的數(shù)據(jù)丟失和損壞。文獻[60]提出一種基于稀疏表示的大量缺失數(shù)據(jù)重建方法,分析并利用感知數(shù)據(jù)的時間穩(wěn)定性、空間相關(guān)性及低秩特性,針對不同的數(shù)據(jù)丟失模式進行分析并獲得了較好的重建精度。而后,文獻[61]進一步完善方法,考慮多參量數(shù)據(jù)間同樣具有較強的相關(guān)性,增加了多參量數(shù)據(jù)約束來利用該特征,進一步提高了重建精度??紤]到WSNs不僅存在丟失,同時還存在異常值,文獻[54]提出了一種基于MC的兩階段數(shù)據(jù)恢復(fù)方案(MC-two-phase),該方案利用MC技術(shù),充分利用環(huán)境數(shù)據(jù)的固有特性,恢復(fù)存在數(shù)據(jù)丟失或損壞問題的數(shù)據(jù)矩陣。為了能在恢復(fù)丟失及異常數(shù)據(jù)的同時檢測出異常節(jié)點的位置,文獻[62]基于環(huán)境數(shù)據(jù)的低階特征,提出一種基于結(jié)構(gòu)噪聲矩陣補全的彈性網(wǎng)絡(luò)正則化數(shù)據(jù)重建算法。
由于基于低秩矩陣模型的WSNs數(shù)據(jù)匯聚及重建方法能夠更有效地利用WSNs數(shù)據(jù)的時空相關(guān)性,近年來相關(guān)成果相繼涌現(xiàn),極大的推動了WSNs數(shù)據(jù)匯聚方法的研究。
4 存在的問題及展望
壓縮感知和低秩理論是近年來信號處理領(lǐng)域的熱點研究,其本質(zhì)是利用數(shù)據(jù)不同階的稀疏性,在WSNs數(shù)據(jù)匯聚研究中也取得了很多具有突破性的進展。一般來講,數(shù)據(jù)匯聚方法的改進主要有3個方面:重建算法的優(yōu)化、變換矩陣的設(shè)計和傳感矩陣的構(gòu)造。然而,針對大規(guī)模、高密度的WSNs數(shù)據(jù)匯聚方法研究仍存在挑戰(zhàn)性。首先,在部分節(jié)點處于休眠狀態(tài)的情況下,如何保證感知到的數(shù)據(jù)準確地傳輸?shù)交荆苊鈹?shù)據(jù)的丟失。其次,為減少網(wǎng)絡(luò)能量消耗采用稀疏采樣進行數(shù)據(jù)匯聚時,如何選擇具有代表性的節(jié)點使得數(shù)據(jù)重建時的準確度更高仍需進一步研究。最后,在數(shù)據(jù)匯聚方案的設(shè)計上,如何防止因能量耗盡或惡意攻擊而可能導(dǎo)致的節(jié)點故障,造成系統(tǒng)的魯棒性差等技術(shù)瓶頸仍需科研人員進一步研究探索。
針對多參量異構(gòu)WSNs,僅基于CS與低秩矩陣雖可有效利用各類信號的時空冗余性,卻忽略了各類型感知信號數(shù)據(jù)間的相關(guān)性。多參量異構(gòu)WSNs中不同節(jié)點,不同采集時刻及不同類型的感知數(shù)據(jù)呈現(xiàn)三維特性,如進一步考慮節(jié)點的空間位置,可呈現(xiàn)更高維特性。此時,所需處理的高維數(shù)據(jù)采用二維矩陣的形式并不足以刻畫其復(fù)雜的內(nèi)在結(jié)構(gòu)。近年來,張量[63]作為向量、矩陣的高階拓展,能直接表示高維數(shù)據(jù),能夠更好地刻畫高維數(shù)據(jù)的內(nèi)在相關(guān)性,對高維數(shù)據(jù)處理具有明顯的優(yōu)勢,受到了研究人員的重點關(guān)注,目前已成功地應(yīng)用于高維信號處理、計算機視覺、機器學(xué)習等領(lǐng)域[64-70],為有效利用多參量異構(gòu)WSNs中數(shù)據(jù)間相關(guān)性提供了一種全新的技術(shù)手段。目前存在多種張量分解形式,典型的有CANDECOMP/PARAFAC (CP)[71]分解,Tucker[72]分解,Tensor Singular Value Decomposition (t-SVD)[73]和Tensor-train (TT)[74]分解等。雖然基于張量的WSNs數(shù)據(jù)感知重建研究相繼提出[75-77],但其相關(guān)研究較少。將張量應(yīng)用于WSNs的數(shù)據(jù)匯聚中時,數(shù)據(jù)主要按照節(jié)點感知數(shù)據(jù)的類型、分布位置和感知時間分布在張量中。基于稀疏表示的無線傳感器網(wǎng)絡(luò)通過部分采樣方式獲取數(shù)據(jù),而稀疏采樣的方式和與之適應(yīng)的匯聚方法密不可分,如何在滿足匯聚方法的前提下設(shè)計能夠捕獲更多數(shù)據(jù)信息的稀疏采樣方式仍需進一步研究。且由于多參量異構(gòu)WSNs數(shù)據(jù)的特殊性,需進一步探索適用于該高維數(shù)據(jù)的張量分解模型,以有效利用各類型感知信號數(shù)據(jù)本身的時空相關(guān)性及其之間的相關(guān)性。
大規(guī)模部署的WSNs中傳感器往往因自然災(zāi)害、節(jié)點自身問題等其他事件的影響,會導(dǎo)致匯聚到的數(shù)據(jù)存在異常和噪聲,針對異常值和噪聲,可基于魯棒主元分析(Robust PCA, RPCA)[78]進行探索,該問題同樣可推廣到張量模型下解決多參量異構(gòu)WSNs中異常值及噪聲問題。同時,WSNs數(shù)據(jù)匯聚也可采用機器學(xué)習方法[79]去除故障節(jié)點、優(yōu)化簇頭選擇等??蒲腥藛T仍需進一步研究,以推動大規(guī)模、高密度、多參量的WSNs的實際部署和高效的數(shù)據(jù)匯聚。
5 結(jié)語
無線傳感器網(wǎng)絡(luò)中,基于稀疏表示的數(shù)據(jù)匯聚方法能夠有效降低網(wǎng)絡(luò)傳輸數(shù)據(jù)量,延長網(wǎng)絡(luò)生命周期。本文主要介紹了基于壓縮感知(約束數(shù)據(jù)一階稀疏性)及低秩矩陣模型(約束數(shù)據(jù)二階稀疏性)的WSNs數(shù)據(jù)匯聚方法。基于WSNs數(shù)據(jù)的時空相關(guān)性,這些方法從不同角度對WSNs數(shù)據(jù)匯聚進行了研究,致力于降低網(wǎng)絡(luò)能耗,提高數(shù)據(jù)重建精度。同時,對當前WSNs數(shù)據(jù)匯聚中存在的問題及后續(xù)研究方向進行了探討與展望,通過研究更高階的張量形式進一步利用各類感知信號間相關(guān)性,切實推動大規(guī)模、高密度、多參量的WSNs的實際部署。
參考文獻:
[1]? ? AKYILDIZ I F,SU W,SANKARASUBRAMANIAM Y,et al. Wireless sensor networks:a survey[J]. Computer Networks,2002,38(4):393-422.
[2]? ? ATZORI L,IERA A,MORABITO G. The Internet of Things:a survey[J]. Computer Networks,2010,54(15):2787-2805.
[3]? ? PRABHU B,PRADEEP M,GAJENDRAN E. Military Applications of Wireless Sensor Network System[J]. Scientific Digest-A Multidisciplinary Journal of Scientific Research & Education,2016,2(12):164-168.
[4]? ? XU G,SHEN W,WANG X. Applications of wireless sensor networks in marine environment monitoring:a survey[J]. Sensors,2014,14(9):16932-16954.
[5]? ? VO M T,THANH NGHI T T,TRAN V S,et al. Wireless sensor network for real time healthcare monitoring:network design and performance evaluation simulation[C]//5th International Conference on Biomedical Engineering in Vietnam,2015.
[6]? ? JAWAD H M,NORDIN R,GHARGHAN S K,et al. Energy-efficient wireless sensor networks for precision agriculture:a review[J]. Sensors,2017,17(8):E1781.
[7]? ? CAND?S E. Compressive sampling[C]//Proceedings of the International Congress of Mathematicians Madrid,August 22–30,2006. Zuerich,Switzerland:European Mathematical Society Publishing House:1433-1452.
[8]? ? DONOHO D L. Compressed sensing[J]. IEEE Transactions on Information Theory,2006,52(4):1289-1306.
[9]? ? CAND?S E J,RECHT B. Exact matrix completion via convex optimization[J]. Foundations of Computational Mathematics,2009,9(6):717-772.
[10]? CANDES E J,PLAN Y. Matrix completion with noise[J]. Proceedings of the IEEE,2010,98(6):925-936.
[11]? MUTHUKRISHNAN S. Data streams:algorithms and applications[J]. Foundations and Trends? in Theoretical Computer Science,2005,1(2):117-236.
[12]? DONOHO D L. For most large underdetermined systems of linear equations the minimal[J]. Communications on Pure and Applied Mathematics:A Journal Issued by the Courant Institute of Mathematical Sciences,2006,59(6):797-829.
[13]? CHEN S S,DONOHO D L,SAUNDERS M A. Atomic decomposition by basis pursuit[J]. SIAM Review,2001,43(1):129-159.
[14]? NEEDELL D,TROPP J A. CoSaMP:Iterative signal recovery from incomplete and inaccurate samples[J]. Applied and Computational Harmonic Analysis,2009,26(3):301-321.
[15]? DAI W,MILENKOVIC O. Subspace pursuit for compressive sensing signal reconstruction[J]. IEEE Transactions on Information Theory,2009,55(5):2230-2249.
[16]? BLUMENSATH T,DAVIES M E. Iterative hard thresholding for compressed sensing[J]. Applied and Computational Harmonic Analysis,2009,27(3):265-274.
[17]? WANG W,WANG D,JIANG Y. Energy efficient distributed compressed data gathering for sensor networks[J]. Ad Hoc Networks,2017,58:112-117.
[18]? LIANG J,MAO C C. Distributed compressive sensing in heterogeneous sensor network[J]. Signal Processing,2016,126:96-102.
[19]? WU X G,XIONG Y,YANG P L,et al. Sparsest random scheduling for compressive data gathering in wireless sensor networks[J]. IEEE Transactions on Wireless Communications,2014,13(10):5867-5877.
[20]? RAN R,OH H. Adaptive sparse random projections for wireless sensor networks with energy harvesting constraints[J]. EURASIP Journal on Wireless Communications and Networking,2015,2015(1):1-10.
[21]? BAJWA W,HAUPT J,SAYEED A,et al. Compressive wireless sensing[C]// 5th International Conference on Information Processing in Sensor Networks. Nashville,TN,USA. IEEE,2006.
[22]? LUO C,WU F,SUN J,et al. Compressive data gathering for large-scale wireless sensor networks[C]// Proceedings of the 15th annual international conference on Mobile computing and networking. 2009:145-156.
[23]? LUO C,WU F,SUN J,et al. Efficient measurement generation and pervasive sparsity for compressive data gathering[J]. IEEE Transactions on Wireless Communications,2010,9(12):3728-3738.
[24]? LUO J,XIANG L,ROSENBERG C. Does compressed sensing improve the throughput of wireless sensor networks?[C]// IEEE International Conference on Communications. May 23-27,2010,Cape Town,South Africa. IEEE,2010:1-6.
[25]? EBRAHIMI D,ASSI C. A distributed method for compressive data gathering in wireless sensor networks[J]. IEEE Communications Letters,2014,18(4):624-627.
[26]? PAKDAMAN TIRANI S,AVOKH A. On the performance of sink placement in WSNs considering energy-balanced compressive sensing-based data aggregation[J]. Journal of Network and Computer Applications,2018,107:38-55.
[27]? LI B,YANG H J,LIU G L,et al. An energy-efficient routing algorithm in three-dimensional underwater sensor networks based on compressed sensing[J]. Information,2017,8(2):66.
[28]? ALAGIRISAMY M,CHOW C O,NOORDIN K A B. Compressive sensing with perceptron based routing for varying traffic intensity based on capsule networks[J]. Computers & Electrical Engineering,2019,79:106446.
[29]? XIANG L,LUO J,DENG C W,et al. Dual-level compressed aggregation:recovering fields of physical quantities from incomplete sensory data[EB/OL]. 2011
[30]? MASOUM A,MERATNIA N,HAVINGA P J M. A distributed compressive sensing technique for data gathering in wireless sensor networks[J]. Procedia Computer Science,2013,21:207-216.
[31]? JIN W,TANG S J,YIN B C,et al. Data gathering in wireless sensor networks? through? intelligent? compressive? sensing[C]//2012 Proceedings IEEE INFOCOM. March 25-30,2012,Orlando,F(xiàn)L,USA. IEEE,2012:603-611.
[32]? XU Y,SUN G L,GENG T,et al. Compressive multi-timeslots data gathering with total variation regularization for wireless sensor networks[J]. IEEE Communications Letters,2019,23(4):648-651.
[33]? SUN P,WU L T,WANG Z B,et al. Sparsest random sampling for cluster-based compressive data gathering in wireless sensor networks[J]. IEEE Access,2018,6:36383-36394.
[34]? PACHARANEY U S,GUPTA R K. Clustering and compressive data gathering in wireless sensor network[J]. Wireless Personal Communications,2019,109(2):1311-1331.
[35]? PACHARANEY U S,GUPTA R K. Cluster restructuring and compressive data gathering for transmission efficient wireless sensor network[C]//Intelligent Communication Technologies and Virtual Mobile Networks. Cham:Springer International Publishing,2019:1-18.
[36]? 吳宣夠,儲昭斌,鄭嘯,等. 鏈路不可靠下稀疏投影無線傳感器網(wǎng)絡(luò)數(shù)據(jù)收集研究[J]. 計算機學(xué)報,2019,42(2):388-402.
[37]? HAO J,ZHANG B X,JIAO Z Z,et al. Adaptive compressive sensing based sample scheduling mechanism for wireless sensor networks[J]. Pervasive and Mobile Computing,2015,22:113-125.
[38]? CHEN S G,LIU J C,WANG K,et al. A hierarchical adaptive spatio-temporal data compression scheme for wireless sensor networks[J]. Wireless Networks,2019,25(1):429-438.
[39]? CHEN J,JIA J,DENG Y S,et al. Adaptive compressive sensing and data recovery for periodical monitoring wireless sensor networks[J]. Sensors,2018,18(10):E3369.
[40]? EBRAHIMI D,ASSI C. Compressive data gathering using random projection for energy efficient wireless sensor networks[J]. Ad Hoc Networks,2014,16:105-119.
[41]? LI X L,TAO X F,MAO G Q. Unbalanced expander based compressive data gathering in clustered wireless sensor networks[J]. IEEE Access,2017,5:7553-7566.
[42]? LI X L,TAO X F,CHEN Z. Spatio-temporal compressive sensing-based data gathering in wireless sensor networks[J]. IEEE Wireless Communications Letters,2018,7(2):198-201.
[43]? ZHANG C,LI O,TONG X,et al. Spatiotemporal data gathering based on compressive sensing in WSNs[J]. IEEE Wireless Communications Letters,2019,8(4):1252-1255.
[44]? 林宙辰,馬毅. 信號與數(shù)據(jù)處理中的低秩模型—理論、算法與應(yīng)用[EB/OL]. [2020-02-19] https://wenku. baidu. com/view/51701da0e45c3b3566
ec8b0f. html.
[45]? KONISHI K,URUMA K,TAKAHASHI T,et al. Iterative partial matrix shrinkage algorithm for matrix rank minimization[J]. Signal Processing,2014,100:124-131.
[46]? 陳蕾,楊庚,陳正宇,等. 基于結(jié)構(gòu)化噪聲矩陣補全的Web服務(wù)QoS預(yù)測[J]. 通信學(xué)報,2015,36(6):53-63.
[47]? 陳蕾,楊庚,陳正宇,等. 基于線性Bregman迭代的結(jié)構(gòu)化噪聲矩陣補全算法[J]. 計算機學(xué)報,2015,38(7):1357-1371.
[48]? CHENG J,JIANG H B,MA X Q,et al. Efficient data collection with sampling in WSNs:making use of matrix completion techniques[C]//2010 IEEE Global Telecommunications Conference GLOBECOM 2010. December 6-10,2010,Miami,F(xiàn)L,USA. IEEE,2010:1-5.
[49]? CHENG J,YE Q,JIANG H B,et al. STCDG:an efficient data gathering algorithm based on matrix completion for wireless sensor networks[J]. IEEE Transactions on Wireless Communications,2013,12(2):850-861.
[50]? HE J F,SUN G L,ZHANG Y,et al. Data recovery in wireless sensor networks with joint matrix completion and sparsity constraints[J]. IEEE Communications Letters,2015,19(12):2230-2233.
[51]? HE J F,SUN G L,LI Z Z,et al. Compressive data gathering with low-rank constraints for Wireless Sensor networks[J]. Signal Processing,2017,131:73-76.
[52]? XIE K,WANG L L,WANG X,et al. Learning from the past:intelligent on-line weather monitoring based on matrix completion[C]//IEEE 34th International Conference on Distributed Computing Systems. June 30-July 3,2014,Madrid,Spain. IEEE,2014:176-185.
[53]? 顧潔. 基于矩陣補全的無線傳感網(wǎng)數(shù)據(jù)收集技術(shù)研究[D]. 南京:南京郵電大學(xué),2018.
[54]? XIE K,NING X P,WANG X,et al. Recover corrupted data in sensor networks:a matrix completion solution[J]. IEEE Transactions on Mobile Computing,2017,16(5):1434-1448.
[55]? TAN J W,LIU W,WANG T,et al. An adaptive collection scheme-based matrix completion for data gathering in energy-harvesting wireless sensor networks[J]. IEEE Access,2019,7:6703-6723.
[56]? KORTAS M,HABACHI O,BOUALLEGUE A,et al. The energy-aware matrix completion-based data gathering scheme for wireless sensor networks[J]. IEEE Access,2020,8:30772-30788.
[57]? MAO X H,QIU K,LI T J,et al. Spatio-temporal signal recovery based on low rank and differential smoothness[J]. IEEE Transactions on Signal Processing,2018,66(23):6281-6296.
[58]? XU Y,SUN G L,GENG T,et al. Low-energy data collection in wireless sensor networks based on matrix completion[J]. Sensors,2019,19(4):945.
[59]? HE J F,ZHANG X Y,ZHOU Y T,et al. A subspace approach to sparse sampling based data gathering in wireless sensor networks[J]. Sensors,2020,20(4):985.
[60]? KONG L H,XIA M Y,LIU X Y,et al. Data loss and reconstruction in sensor networks[C]//Proceedings IEEE INFOCOM. April 14-19,2013,Turin,Italy. IEEE,2013:1654-1662.
[61]? KONG L H,XIA M Y,LIU X Y,et al. Data loss and reconstruction in wireless sensor networks[J]. IEEE Transactions on Parallel and Distributed Systems,2014,25(11):2818-2828.
[62]? CHEN Z Y,CHEN L,HU G B,et al. Data reconstruction in wireless sensor networks from incomplete and erroneous observations[J]. IEEE Access,2018,6:45493-45503.
[63]? KOLDA T G,BADER B W. Tensor decompositions and applications[J]. SIAM Review,2009,51(3):455-500.
[64]? LU C Y,F(xiàn)ENG J S,CHEN Y D,et al. Tensor robust principal component analysis with a new tensor nuclear norm[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2020,42(4):925-938.
[65]? LIU Y P,CHEN L X,ZHU C. Improved robust tensor principal component analysis via low-rank core matrix[J]. IEEE Journal of Selected Topics in Signal Processing,2018,12(6):1378-1389.
[66]? CHEN Y,HUANG T Z,ZHAO X L. Destriping of multispectral remote sensing image using low-rank tensor decomposition[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2018,11(12):4950-4967.
[67]? GAO S Q,F(xiàn)AN Q B. A mixture of nuclear norm and matrix factorization for tensor completion[J]. Journal of Scientific Computing,2018,75(1):43-64.
[68]? JI T Y,HUANG T Z,ZHAO X L,et al. A non-convex tensor rank approximation for tensor completion[J]. Applied Mathematical Modelling,2017,48:410-422.
[69]? JIANG T X,HUANG T Z,ZHAO X L,et al. Matrix factorization for low-rank tensor completion using framelet prior[J]. Information Sciences,2018,436/437:403-417.
[70]? WANG Y T,ZHAO X L,JIANG T X,et al. A total variation and group sparsity based tensor optimization model for video rain streak removal[J]. Signal Processing:Image Communication,2019,73:96-108.
[71]? HARSHMAN R A. Foundations of the PARAFAC procedure:Model and conditions for an explanatory multi-mode factor analysis[J]. UCLA Working Papers in Phonetics,1970,16:1-84.
[72]? TUCKER L R. The extension of factor analysis to three-dimensional matrices[C]//In Frederiksen. Proceedings of the Contributions to Mathematical Psychology. Holt,Rimehat and Winston,New York,1964:279-311.
[73]? ZHANG Z M,AERON S. Exact tensor completion using t-SVD[J]. IEEE Transactions on Signal Processing,2017,65(6):1511-1526.
[74]? OSELEDETS I V. Tensor-train decomposition[J]. SIAM Journal on Scientific Computing,2011,33(5):2295-2317.
[75]? HE J F,ZHOU Y T,SUN G L,et al. Multi-attribute data recovery in wireless sensor networks with joint sparsity and low-rank constraints based on tensor completion[J]. IEEE Access,2019,7:135220-135230.
[76]? MILOSEVIC B,YANG J,VERMA N,et al. Efficient energy management and data recovery in sensor networks using latent variables based tensor factorization[C]//Proceedings of the 16th ACM international conference on Modeling,analysis & simulation of wireless and mobile systems. Barcelona Spain. New York,NY,USA:ACM,2013.
[77]? HE J F,SUN G L,ZHANG Y,et al. Data recovery in heterogeneous wireless sensor networks based on low-rank tensors[C]//IEEE Symposium on Computers and Communication (ISCC). June 27-30,2016,Messina,Italy. IEEE,2016:616-620.
[78]? CHANDRASEKARAN V,SANGHAVI S,PARRILO P A,et al. Sparse and low-rank matrix decompositions[J]. IFAC Proceedings Volumes,2009,42(10):1493-1498.
[79]? ALSHEIKH M A,LIN S W,NIYATO D,et al. Machine learning in wireless sensor networks:algorithms,strategies,and applications[J]. IEEE Communications Surveys & Tutorials,2014,16(4):1996-2018.