摘" 要:農(nóng)業(yè)不僅是國民經(jīng)濟建設(shè)與發(fā)展的基礎(chǔ),也是社會有序運行的保障。然而每年由于農(nóng)作物病蟲害造成的損失巨大,因此及時精準(zhǔn)地檢測農(nóng)作物病蟲害情況并采取相應(yīng)措施,對于農(nóng)業(yè)發(fā)展有著重要意義。近年來,深度學(xué)習(xí)在圖像識別方面取得巨大進(jìn)展,其中卷積神經(jīng)網(wǎng)絡(luò)具有較好的圖像識別能力,利用該技術(shù)可以準(zhǔn)確地識別農(nóng)作物病蟲害,以便及時地進(jìn)行防治。首先,該文分別綜述農(nóng)作物病蟲害識別的傳統(tǒng)方法、機器學(xué)習(xí)方法、深度學(xué)習(xí)方法,并分析比較3種方法的優(yōu)缺點。其次,闡述國內(nèi)外專家學(xué)者在農(nóng)作物病蟲害識別關(guān)鍵技術(shù)上的研究,分別分析總結(jié)數(shù)據(jù)集的獲取途徑和規(guī)模、數(shù)據(jù)集多種預(yù)處理技術(shù)的作用、數(shù)據(jù)集增強技術(shù)的多種方法、網(wǎng)絡(luò)模型的遷移學(xué)習(xí)和預(yù)處理的作用、網(wǎng)絡(luò)模型的種類和優(yōu)缺點及網(wǎng)絡(luò)模型多種優(yōu)化技術(shù)的特點和優(yōu)缺點。最后,指出目前基于卷積神經(jīng)網(wǎng)絡(luò)的農(nóng)作物病蟲害識別研究的熱點難點,并對其應(yīng)用前景進(jìn)行展望。
關(guān)鍵詞:深度學(xué)習(xí);卷積神經(jīng)網(wǎng)絡(luò);圖像識別;關(guān)鍵技術(shù);病蟲害識別
中圖分類號:TP183" " " 文獻(xiàn)標(biāo)志碼:A" " " " " 文章編號:2096-9902(2024)17-0039-07
Abstract: Agriculture is not only the basis of national economic construction and development, but also the guarantee of social orderly operation. However, due to the huge losses caused by crop diseases and insect pests every year, it is of great significance for agricultural development to detect crop diseases and insect pests timely and accurately and take corresponding measures. In recent years, deep learning has made great progress in image recognition, in which Convolutional Neural Network has a good ability of image recognition, using this technology can accurately identify crop diseases and insect pests for timely prevention and control. First of all, this paper summarizes the traditional methods, machine learning methods and deep learning methods of crop pest identification, and analyzes and compares the advantages and disadvantages of the three methods. Secondly, The research of experts and scholars at home and abroad on the key technologies of crop disease and pest identification are described. This paper analyzes and summarizes the ways and scale of obtaining data sets, the functions of various preprocessing techniques of data sets, various methods of data set enhancement, the role of transfer learning and preprocessing of network models, the types and advantages and disadvantages of network models, and the characteristics, advantages and disadvantages of various optimization techniques of network models. Finally, the hot spots and difficulties of crop pest identification based on Convolutional Neural Network are pointed out, and its application prospect is prospected.
Keywords: deep learning; Convolutional Neural Network; image recognition; key technology; pest identification
國家的經(jīng)濟發(fā)展離不開農(nóng)業(yè)的繁榮。然而,農(nóng)作物的病蟲害問題一直以來都是制約農(nóng)業(yè)發(fā)展的關(guān)鍵因素[1-2]。這些病蟲害嚴(yán)重影響著農(nóng)作物的產(chǎn)量和質(zhì)量,災(zāi)難性病蟲害的發(fā)生導(dǎo)致糧食供應(yīng)短缺。針對農(nóng)作物病蟲害的預(yù)防與治理,傳統(tǒng)的植物病蟲害識別方法需要耗費大量的時間與高昂的費用進(jìn)行人工觀察與專業(yè)鑒定[3]。對于這種情況,人們開始尋求新方法,于是,機器學(xué)習(xí)逐漸被人們重視,利用機器學(xué)習(xí)技術(shù)不僅可以有效減少人工觀察時間,而且可以減少用于專業(yè)鑒定的高昂費用。但是,隨著社會的發(fā)展及人們?nèi)找嬖鲩L的需求,傳統(tǒng)機器學(xué)習(xí)的缺點逐漸顯現(xiàn)出來,其中圖片的特征提取需要手動操作,過程十分繁瑣并且影響精確度,從而導(dǎo)致機器學(xué)習(xí)的算法識別結(jié)果不盡如人意。隨后,深度學(xué)習(xí)迅猛發(fā)展[4],在數(shù)字圖像處理領(lǐng)域取得了突破,遠(yuǎn)遠(yuǎn)優(yōu)于傳統(tǒng)方法[5]。深度學(xué)習(xí)方法主要使用卷積神經(jīng)網(wǎng)絡(luò)模型,這一網(wǎng)絡(luò)模型在大規(guī)模識別任務(wù)中已經(jīng)展現(xiàn)出優(yōu)于相關(guān)專家的識別準(zhǔn)確度[6]。然而,隨著深度學(xué)習(xí)的快速發(fā)展,網(wǎng)絡(luò)結(jié)構(gòu)不斷優(yōu)化?;诰矸e神經(jīng)網(wǎng)絡(luò)的植物病蟲害識別研究具備識別準(zhǔn)確性高、魯棒性強、泛化性好等特點[7]。盡管如此,仍存在一些挑戰(zhàn),例如需要大量的數(shù)據(jù)集提高識別準(zhǔn)確率,大量實驗來確定最優(yōu)的網(wǎng)絡(luò)結(jié)構(gòu)等[8]。為了深入探索卷積神經(jīng)網(wǎng)絡(luò)在農(nóng)業(yè)病蟲害識別領(lǐng)域的研究,本文綜述了農(nóng)業(yè)病蟲害識別的幾種方法,分析了國內(nèi)外專家學(xué)者在基于卷積神經(jīng)網(wǎng)絡(luò)的病蟲害識別關(guān)鍵技術(shù)上的研究進(jìn)展,探討了目前基于卷積神經(jīng)網(wǎng)絡(luò)的農(nóng)業(yè)病蟲害識別研究的熱點和難點,并對其應(yīng)用前景進(jìn)行了展望。
1" 農(nóng)作物病蟲害識別方法
1.1" 傳統(tǒng)農(nóng)作物病蟲害識別方法
傳統(tǒng)人工農(nóng)作物病蟲害識別方法是在長期的農(nóng)業(yè)生產(chǎn)實踐中形成的一類識別技術(shù),主要依賴于農(nóng)業(yè)從業(yè)者的經(jīng)驗和專業(yè)知識,通過人工視覺觀察,來分析判斷植物是否感染病蟲害。
人工農(nóng)作物病蟲害識別方法包括個人經(jīng)驗判斷和基于專家知識的判別,其中,個人依靠長期的經(jīng)驗判斷病蟲害的種類,而專業(yè)的農(nóng)業(yè)病蟲害專家通過專業(yè)的知識判斷病蟲害種類。這些方法對于一些特定病蟲害的識別有一定的準(zhǔn)確性,但遇到與別的病蟲害相似的癥狀,就會受到個人主觀限制,不僅耗時耗力,而且判斷的結(jié)果存在不穩(wěn)定[9]。圖1為傳統(tǒng)農(nóng)作物病蟲害防治流程圖。
1.2" 基于機器學(xué)習(xí)的農(nóng)作物病蟲害識別方法
基于機器學(xué)習(xí)的農(nóng)作物病蟲害識別技術(shù)是利用計算機視覺和人工智能的方法,通過對大量圖像數(shù)據(jù)的學(xué)習(xí)和訓(xùn)練,實現(xiàn)自動化的病蟲害識別。其主要流程包括數(shù)據(jù)采集、數(shù)據(jù)預(yù)處理、特征提取和模型訓(xùn)練[10]。相較于傳統(tǒng)的人工農(nóng)作物病蟲害識別方法,基于機器學(xué)習(xí)的農(nóng)作物病蟲害識別具有客觀性、高效性。然而,基于機器學(xué)習(xí)的農(nóng)作物病蟲害識別技術(shù)也存在一些缺點,該技術(shù)需要專門的人員提前進(jìn)行特征采集,然后傳輸?shù)缴窠?jīng)網(wǎng)絡(luò)中進(jìn)行識別分類。這一過程相當(dāng)復(fù)雜且耗費大量人力物力。圖2為基于機器學(xué)習(xí)的農(nóng)作物病蟲害識別流程圖。
1.3" 基于深度學(xué)習(xí)的農(nóng)作物病蟲害識別技術(shù)
深度學(xué)習(xí)是機器學(xué)習(xí)領(lǐng)域的一個重要分支,其核心特點是利用多層神經(jīng)網(wǎng)絡(luò)進(jìn)行特征學(xué)習(xí)和模式識別。通過多層次的非線性轉(zhuǎn)換,深度學(xué)習(xí)技術(shù)可以從原始數(shù)據(jù)中提取更高級、更抽象的特征,從而更好地區(qū)分不同的病蟲害樣本,提高識別準(zhǔn)確性和泛化能力,在病蟲害識別任務(wù)中,基于深度學(xué)習(xí)的技術(shù)主要采用卷積神經(jīng)網(wǎng)絡(luò)(CNN)和循環(huán)神經(jīng)網(wǎng)絡(luò)(RNN)等模型[11-12]。雖然基于深度學(xué)習(xí)的農(nóng)作物病蟲害識別技術(shù)與機器學(xué)習(xí)的方法在流程上有相似之處,但在特征提取階段,深度學(xué)習(xí)具有明顯優(yōu)勢。深度學(xué)習(xí)可以自動學(xué)習(xí)圖像中的特征,而不需要手動設(shè)計特征。這使得基于深度學(xué)習(xí)的農(nóng)作物病蟲害識別方法在處理大規(guī)模和復(fù)雜數(shù)據(jù)時表現(xiàn)更加優(yōu)越。因此近些年來,深度學(xué)習(xí)中有些網(wǎng)絡(luò)模型在農(nóng)作物病蟲害識別任務(wù)中,已能實現(xiàn)遠(yuǎn)遠(yuǎn)優(yōu)于相關(guān)專家的識別準(zhǔn)確度[13]。圖3為基于深度學(xué)習(xí)的農(nóng)作物病蟲害識別流程圖。
2" 農(nóng)作物病蟲害識別關(guān)鍵技術(shù)
基于卷積神經(jīng)網(wǎng)絡(luò)的農(nóng)作物病蟲害識別關(guān)鍵技術(shù)是一種利用深度學(xué)習(xí)方法自動檢測和識別農(nóng)作物病蟲害的先進(jìn)技術(shù)。其主要內(nèi)容有數(shù)據(jù)集獲取、數(shù)據(jù)預(yù)處理、數(shù)據(jù)增強、遷移學(xué)習(xí)和預(yù)訓(xùn)練、神經(jīng)網(wǎng)絡(luò)模型選擇和網(wǎng)絡(luò)模型優(yōu)化等[14]。
2.1" 病蟲害數(shù)據(jù)集的獲取技術(shù)
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,適當(dāng)?shù)臄?shù)據(jù)準(zhǔn)備可以幫助模型找到最優(yōu)參數(shù),提高識別性能。在農(nóng)業(yè)病蟲害識別領(lǐng)域,數(shù)據(jù)源主要可以通過3種方式獲得:第一種是利用相關(guān)研究機構(gòu)提供的數(shù)據(jù)或現(xiàn)有的公共數(shù)據(jù)集,比如Plant Village、AI Challenger等;第二種是研究人員自己搜集,比如利用手機、相機等設(shè)備進(jìn)行圖片采集;第三種則是從互聯(lián)網(wǎng)上搜索圖片。通過這些數(shù)據(jù)獲取方式為卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別技術(shù)提供了豐富的數(shù)據(jù)資源,有助于推動該領(lǐng)域的研究和應(yīng)用發(fā)展。本文總計所研究文章的數(shù)據(jù)集選擇情況見表1。
由表1可知,研究者們廣泛使用開源數(shù)據(jù)集作為訓(xùn)練和測試模型的主要數(shù)據(jù)來源。部分研究者選擇通過使用手機或相機自行構(gòu)建數(shù)據(jù)集。相對而言,采用互聯(lián)網(wǎng)搜索、網(wǎng)絡(luò)爬蟲獲取數(shù)據(jù)集的研究者數(shù)量較少。
2.2" 病蟲害數(shù)據(jù)集的預(yù)處理技術(shù)
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,病蟲害數(shù)據(jù)集的預(yù)處理技術(shù)在農(nóng)業(yè)科學(xué)領(lǐng)域具有重要意義。在采集的過程中,病蟲害圖像很容易受到無關(guān)信息的干擾,比如灰塵、雜物、光照和模糊等,從而會影響識別的準(zhǔn)確度,所以需要對圖像數(shù)據(jù)集進(jìn)行除噪除雜等預(yù)處理,以確保圖像質(zhì)量。另外需要對圖像進(jìn)行尺寸標(biāo)準(zhǔn)化,通常將它們調(diào)整為相同的大小,以便于神經(jīng)網(wǎng)絡(luò)的處理。具體預(yù)處理方式及其說明見表2。
由表2可知,采用調(diào)整圖像尺寸、灰度化、PCA白化、轉(zhuǎn)變顏色模型和降噪等多種數(shù)據(jù)預(yù)處理方式,研究者在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別領(lǐng)域取得了顯著進(jìn)展。這些技術(shù)的綜合應(yīng)用為該領(lǐng)域的研究和實際應(yīng)用提供了有力的支持,推動了病蟲害識別技術(shù)的不斷發(fā)展與優(yōu)化。
2.3" 病蟲害數(shù)據(jù)集的增強技術(shù)
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,數(shù)據(jù)增強被廣泛采用,以提高模型的魯棒性和泛化能力。研究者運用不同類型的數(shù)據(jù)增強方法來增加訓(xùn)練數(shù)據(jù)的多樣性,如幾何變換類、空間變換類、顏色變換類等。文章所涉及數(shù)據(jù)增強方式見表3。
由表3可知,在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別中的研究者廣泛應(yīng)用各類數(shù)據(jù)增強方法。這些數(shù)據(jù)增強技術(shù)能夠有效地增加數(shù)據(jù)集的多樣性,提高模型的泛化能力和魯棒性,進(jìn)而推動病蟲害識別技術(shù)的發(fā)展和應(yīng)用。通過綜合運用這些方法為卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的精度和可靠性帶來了積極的影響,為農(nóng)業(yè)保護(hù)和食品安全領(lǐng)域提供了有力的支持。
2.4" 網(wǎng)絡(luò)模型的遷移學(xué)習(xí)和預(yù)訓(xùn)練技術(shù)
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,許多任務(wù)面臨著數(shù)據(jù)不足、類別不平衡、領(lǐng)域偏移等挑戰(zhàn),導(dǎo)致模型難以泛化到新的、未見過的數(shù)據(jù)。為了應(yīng)對這些挑戰(zhàn)并提高深度學(xué)習(xí)的效率和性能,許多研究人員選擇了遷移學(xué)習(xí)的方法來訓(xùn)練模型。遷移學(xué)習(xí)的原理是將訓(xùn)練好的參數(shù)應(yīng)用在別的神經(jīng)網(wǎng)絡(luò)模型中,這樣就可以很快地進(jìn)行模型訓(xùn)練,極大地節(jié)約了資源和時間。遷移學(xué)習(xí)的過程如圖4所示。
總的來說,遷移學(xué)習(xí)和預(yù)訓(xùn)練是神經(jīng)網(wǎng)絡(luò)中的2種強大的技術(shù),它們能夠提高模型的泛化能力和訓(xùn)練速度,并在數(shù)據(jù)稀缺或復(fù)雜任務(wù)的場景下發(fā)揮重要作用。
2.5" 神經(jīng)網(wǎng)絡(luò)模型的選擇
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別的研究中,網(wǎng)絡(luò)結(jié)構(gòu)模型的選擇是一個至關(guān)重要的決策,因為不同的網(wǎng)絡(luò)結(jié)構(gòu)會直接影響模型的性能和學(xué)習(xí)能力。本文深入研究了多種網(wǎng)絡(luò)模型,如表4中的AlexNet、VGGNet、ResNet、EfficientNet、MobileNet和Inception。
由表4可知,每種網(wǎng)絡(luò)結(jié)構(gòu)模型在病蟲害識別任務(wù)中都展現(xiàn)出了各自的優(yōu)勢和局限性。研究者們根據(jù)不同的任務(wù)需求和資源限制,選擇合適的網(wǎng)絡(luò)結(jié)構(gòu)模型,以獲得最佳的識別性能。未來的研究可以進(jìn)一步優(yōu)化和改進(jìn)這些網(wǎng)絡(luò)結(jié)構(gòu),以提升病蟲害識別技術(shù)的效率和準(zhǔn)確度,推動該領(lǐng)域的發(fā)展和應(yīng)用。
2.6" 神經(jīng)網(wǎng)絡(luò)模型的優(yōu)化技術(shù)
在卷積神經(jīng)網(wǎng)絡(luò)的演進(jìn)過程中,優(yōu)化網(wǎng)絡(luò)模型的方法成為一個至關(guān)重要的研究領(lǐng)域。研究者們提出了多種優(yōu)化方法。從最基礎(chǔ)的隨機梯度下降(SGD)到更加高級的優(yōu)化器,如Adam、Adamax等。這些優(yōu)化算法能夠自適應(yīng)地調(diào)整學(xué)習(xí)率和梯度更新策略,有效幫助網(wǎng)絡(luò)更快地收斂到最優(yōu)解,從而加速訓(xùn)練過程。常見優(yōu)化器算法見表5。
由表5可知,選擇優(yōu)化器算法時,研究者們需考慮數(shù)據(jù)集規(guī)模、網(wǎng)絡(luò)結(jié)構(gòu)復(fù)雜性、計算資源等因素,結(jié)合實際情況選擇合適的優(yōu)化器來提升模型的收斂速度和性能。因此仔細(xì)選擇和調(diào)整更合適的優(yōu)化器能夠為病蟲害識別等任務(wù)帶來更好的效果。
3" 問題與展望
隨著深度學(xué)習(xí)技術(shù)的蓬勃發(fā)展,利用卷積神經(jīng)網(wǎng)絡(luò)(CNN)自動提取圖像特征,已為當(dāng)前最炙手可熱的深度學(xué)習(xí)技術(shù)之一,但其在農(nóng)作物病蟲害識別方面的應(yīng)用尚處于初級階段,仍面臨一定程度的挑戰(zhàn),亟待深入挖掘和拓展。
3.1" 存在的問題
3.1.1" 數(shù)據(jù)收集問題
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別中,數(shù)據(jù)收集困難是一個重要的問題。因為農(nóng)業(yè)病蟲害的數(shù)據(jù)采集受到多種因素的限制,如不同地理環(huán)境、氣候條件、農(nóng)業(yè)生產(chǎn)區(qū)域的分散性等,導(dǎo)致研究人員難以獲得全面而多樣化的數(shù)據(jù)樣本。
3.1.2" 圖像檢測問題
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別中,圖像檢測是一個具有挑戰(zhàn)性的任務(wù)。環(huán)境復(fù)雜、病蟲害多樣、形態(tài)不規(guī)則和背景干擾多,增加了農(nóng)作物病蟲害圖像檢測的難度。
3.1.3" 模型訓(xùn)練和硬件資源問題
卷積神經(jīng)網(wǎng)絡(luò)的深度結(jié)構(gòu)可以提高準(zhǔn)確率,但網(wǎng)絡(luò)結(jié)構(gòu)深度越深所需要訓(xùn)練時間越長,以及硬件資源要求越高。在嵌入式平臺上,GPU性能遠(yuǎn)低于計算機,從而導(dǎo)致卷積神經(jīng)網(wǎng)絡(luò)在嵌入式平臺上實時目標(biāo)檢測速度明顯下降,這給實時目標(biāo)檢測的應(yīng)用提出了挑戰(zhàn)。
3.2" 未來展望
在卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別領(lǐng)域,會面臨數(shù)據(jù)獲取、圖像檢測、模型訓(xùn)練和硬件資源等問題的挑戰(zhàn)。然而,隨著科學(xué)技術(shù)的不斷進(jìn)步和創(chuàng)新,這些困難均可克服,并可推動該領(lǐng)域取得更為顯著的發(fā)展。首先,對于數(shù)據(jù)獲取困難問題,未來的研究可以通過協(xié)調(diào)國內(nèi)外研究人員的合作,實現(xiàn)數(shù)據(jù)的共享與交換,同時,可以探索利用新興技術(shù),如無人機和遙感等,來構(gòu)建更大規(guī)模病蟲害數(shù)據(jù)庫。其次,針對圖像檢測困難問題,未來的研究可以聚焦于改進(jìn)圖像預(yù)處理和增強技術(shù),以提高圖像質(zhì)量和準(zhǔn)確性。再次,針對模型訓(xùn)練和硬件資源問題,未來的研究可以致力于開發(fā)更加優(yōu)化的神經(jīng)網(wǎng)絡(luò)模型,比如設(shè)計輕量級網(wǎng)絡(luò)結(jié)構(gòu)、引入注意力機制、模型壓縮等技術(shù),以提高訓(xùn)練效率。最后,針對嵌入式平臺的硬件資源限制,可以探索深度學(xué)習(xí)加速器和模型優(yōu)化技術(shù),以提高神經(jīng)網(wǎng)絡(luò)在嵌入式設(shè)備上的實時檢測性能。總之,神經(jīng)網(wǎng)絡(luò)病蟲害識別領(lǐng)域面臨著一系列挑戰(zhàn),但隨著科技的不斷進(jìn)步,這些困難將逐步得到解決。未來的研究將以構(gòu)建更大規(guī)模的數(shù)據(jù)集、優(yōu)化模型訓(xùn)練、利用硬件資源等為重點,推動卷積神經(jīng)網(wǎng)絡(luò)病蟲害識別技術(shù)在農(nóng)業(yè)生產(chǎn)中的應(yīng)用。
4" 結(jié)論
本文綜述了國內(nèi)外研究人員利用神經(jīng)網(wǎng)絡(luò)模型進(jìn)行病蟲害識別技術(shù)的研究進(jìn)展,討論了病蟲害識別關(guān)鍵技術(shù)。研究表明,使用更豐富、真實且多樣化的數(shù)據(jù)源,并通過數(shù)據(jù)預(yù)處理和遷移學(xué)習(xí),可以促使病蟲害識別的準(zhǔn)確率得到很好的提升。尤其選擇合適的神經(jīng)網(wǎng)絡(luò)模型,并通過模型的改進(jìn)和優(yōu)化,能夠促使特定的任務(wù)得到更高效的解決。然而,在探討現(xiàn)階段技術(shù)不足和未來發(fā)展趨勢時發(fā)現(xiàn),農(nóng)業(yè)病蟲害識別領(lǐng)域在卷積神經(jīng)網(wǎng)絡(luò)應(yīng)用方面仍面臨一些挑戰(zhàn)。短期內(nèi),卷積神經(jīng)網(wǎng)絡(luò)在農(nóng)業(yè)病蟲害識別上可能無法徹底解決問題。但是值得肯定的是,現(xiàn)階段的很多經(jīng)驗在未來將成為背后大數(shù)據(jù)的重要組成部分。因此,在未來的發(fā)展中,對卷積神經(jīng)網(wǎng)絡(luò)在農(nóng)業(yè)領(lǐng)域的光明前景將持樂觀態(tài)度,并期待通過不斷地研究和探索,推動該技術(shù)的進(jìn)步與發(fā)展,使其成為農(nóng)業(yè)病蟲害防控的有力工具。
參考文獻(xiàn):
[1] GERALD A,CARLSON.A decision theoretic Approach to crop disease prediction and control[J]. American Journal of Agricultural Economics,1970,52(2):216-223.
[2] AL-HIARY H, BANI-AHMAD S, REYALAT M, et al. Fast and accurate detection and classification of plant disease[J].International Journal of Computer Applications, 2011,17(1):31-38.
[3] LI Y, NIE J, CHAO X. Do we really need deep CNN for plant diseases identification?[J]. Computers and Electronics in Agriculture,2020,178(3):105803.
[4] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. nature, 2015, 521(7553): 436-444.
[5] LIU J, WANG X. Plant diseases and pests detection based on deep learning: a review[J]. Plant Methods,2021,17:1-18.
[6] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems,2012,25.
[7] 駱潤玫,王衛(wèi)星.基于卷積神經(jīng)網(wǎng)絡(luò)的植物病蟲害識別研究綜述[J].自動化與信息工程,2021,42(5):1-10.
[8] JIANG F, LU Y, CHEN Y, et al. Image recognition of four rice leaf diseases based on deep learning and support vector machine[J]. Computers and Electronics in Agriculture,2020, 179(2):105824.
[9] NGUGI L C, ABELWAHAB M, ABO-ZAHHAD M. Recent advances in image processing techniques for automated leaf pest and disease recognition-A review[J]. Information processing in agriculture,2021,8(1):27-51.
[10] 孫成會,薛凱鑫.基于人工智能的圖像識別技術(shù)分析[J].電子測試,2020(16):139-140.
[11] GOODFELLOW I,BENGIO Y,COURVILLE A.Deep learning[M].MIT press,2016.
[12] GU J, WANG Z, KUEN J, et al. Recent advances in convolutional neural networks[J].Pattern recognition,2018, 77:354-377.
[13] BOULENT J, FOUCHER S, TH?魪AU J, et al. Convolutional neural networks for the automatic identification of plant diseases[J].Frontiers in plant science,2019,10:941.
[14] 翟肇裕,曹益飛,徐煥良,等.農(nóng)作物病蟲害識別關(guān)鍵技術(shù)研究綜述[J].農(nóng)業(yè)機械學(xué)報,2021,52(7):1-18.
[15] T?譈RKO■LU M, HANBAY D. Plant disease and pest detection using deep learning-based features[J]. Turkish Journal of Electrical Engineering and Computer Sciences,2019,27(3):1636-1651.
[16] KAUR S, PANDEY S, GOEL S. Semi-automatic leaf disease detection and classification system for soybean culture[J].IET Image Processing,2018,12(6):1038-1048.
[17] LIU B, TAN C, LI S, et al. A data augmentation method based on generative adversarial networks for grape leaf disease identification[J]. IEEE Access,2020,8:102188-102198.
[18] DARSHAN V S. Automated Diagnosis and Cataloguing of Foliar Disease in Apple Treesusing Ensemble of Deep Neural Networks[Z].International Research Journal of Engineering and Technology(IRJET),2020.
[19] NAGI R, TRIPATHY S S. Deep convolutional neural network based disease identification in grapevine leaf images[J].Multimedia Tools and Applications,2022,81(18):24995-25006.
[20] JI M, ZHANG L, WU Q. Automatic grape leaf diseases identification via UnitedModel based on multiple convolutional neural networks[J]. Information Processing in Agriculture,2020,7(3):418-426.
[21] CHEN J, CHEN J, ZHANG D, et al. Using deep transfer learning for image-based plant disease identification[J]. Computers and Electronics in Agriculture,2020,173:105393.
[22] MALVADE N N, YAKKUNDIMATH R, SAUNSHI G, et al. A comparative analysis of paddy crop biotic stress classification using pre-trained deep neural networks[J]. Artificial Intelligence in Agriculture,2022,6:167-175.
[23] ZHANG X, QIAO Y, MENG F, et al. Identification of maize leaf diseases using improved deep convolutional neural networks[J].IEEE Access, 2018,6:30370-30377.
[24] WANG P, NIU T, MAO Y, et al. Identification of apple leaf diseases by improved deep convolutional neural networks with an attention mechanism[J].Frontiers in Plant Science,2021,12:723294.
[25] MAEDA-GUTI?魪RREZ V, GALV?魣N-TEJADA C E, ZANELLA-CALZADA L A, et al. Comparison of convolutional neural network architectures for classification of tomato plant diseases[J].Applied Sciences,2020,10(4): 1245.
[26] BIR P, KUMAR R, SINGH G. Transfer learning based tomato leaf disease detection for mobile applications[C]//2020 IEEE International Conference on Computing, Power and Communication Technologies (GUCON). IEEE,2020: 34-39.
[27] BRAHIMI M, BOUKHALFA K, MOUSSAOUI A. Deep learning for tomato diseases: classification and symptoms visualization[J].Applied Artificial Intelligence,2017,31(4):299-315.
[28] JIANG P, CHEN Y, LIU B, et al. Real-time detection of apple leaf diseases using deep learning approach based on improved convolutional neural networks[J].IEEE Access, 2019,7:59069-59080.
[29] LIU B, ZHANG Y, HE D J, et al. Identification of apple leaf diseases based on deep convolutional neural networks[J]. Symmetry,2017,10(1):11.
[30] GEETHARAMANI G, PANDIAN A. Identification of plant leaf diseases using a nine-layer deep convolutional neural network[J].Computers amp; Electrical Engineering,2019,76: 323-338.
[31] MOHANTY S P, HUGHES D P, SALATH?魪 M. Using deep learning for image-based plant disease detection[J]. Frontiers in plant science,2016,7:1419.
[32] ?譈MIT A ,MURAT U ,KEMAL A, et al. Plant leaf disease classification using EfficientNet deep learning model[J]. Ecological Informatics,2021,61:101182.
[33] ARSENOVIC M, KARANOVIC M, SLADOJEVIC S, et al. Solving current limitations of deep learning based approaches for plant disease detection[J].Symmetry,2019,11(7):939.
[34] SALEEM M H, POTGIETER J, ARIF K M. Plant disease classification: A comparative evaluation of convolutional neural networks and deep learning optimizers[J]. Plants, 2020,9(10):1319.
[35] CHAO X, SUN G, ZHAO H, et al. Identification of apple tree leaf diseases based on deep learning models[J]. Symmetry,2020,12(7):1065.
[36] YAKKUNDIMATH R, SAUNSHI G, ANAMI B, et al. Classification of rice diseases using convolutional neural network models[J]. Journal of The Institution of Engineers (India): Series B,2022,103(4):1047-1059.
[37] SINGH P, VERMA A, ALEX J S R. Disease and pest infection detection in coconut tree through deep learning techniques[J]. Computers and electronics in agriculture,2021, 182:105986.
[38] ALTUNTAS Y, C?魻MERT Z, KOCAMAZ A F. Identification of haploid and diploid maize seeds using convolutional neural networks and a transfer learning approach[J]. Computers and Electronics in Agriculture,2019,163:104874.
[39] NASIRI A, TAHERI-GARAVAND A, ZHANG Y D. Image-based deep learning automated sorting of date fruit[J]. Postharvest biology and technology,2019,153:133-141.
[40] FUENTES A, YOON S, PARK D S. Deep learning-based phenotyping system with glocal description of plant anomalies and symptoms[J].Frontiers in Plant Science,2019,10: 1321.
[41] 曾偉輝, 李淼, 張健, 等. 面向農(nóng)作物病害識別的高階殘差卷積神經(jīng)網(wǎng)絡(luò)研究[J]. 中國科學(xué)技術(shù)大學(xué)學(xué)報,2019, 49(10):781-790.
[42] PICON A, SEITZ M, ALVAREZ-GILA A, et al. Crop conditional Convolutional Neural Networks for massive multi-crop plant disease classification over cell phone acquired images taken on real field conditions[J]. Computers and Electronics in Agriculture,2019,167:105093.
[43] WAHEED A, GOYAL M, GUPTA D, et al. An optimized dense convolutional neural network model for disease recognition and classification in corn leaf[J]. Computers and Electronics in Agriculture,2020,175:105456.
[44] RAMCHARAN A, BARANOWSKI K, MCCLOSKEY P, et al. Deep learning for image-based cassava disease detection[J]. Frontiers in plant science,2017,8:1852.
[45] SLADOJEVIC S, ARSENOVIC M, ANDERLA A, et al. Deep neural networks based recognition of plant diseases by leaf image classification[J]. Computational intelligence and neuroscience,2016,6:1-11.
[46] KUKREJA V, DHIMAN P. A Deep Neural Network based disease detection scheme for Citrus fruits[C]//2020 International conference on smart electronics and communication (ICOSEC).IEEE,2020:97-101.
[47] LU Y, YI S, ZENG N, et al. Identification of rice diseases using deep convolutional neural networks[J]. Neurocomputing,2017,267:378-384.
[48] ZHANG S, ZHANG S, ZHANG C, et al. Cucumber leaf disease identification with global pooling dilated convolutional neural network[J]. Computers and Electronics in Agriculture,2019,162:422-430.
[49] DUARTE-CARVAJALINO J M, ALZATE D F, RAMIREZ A A, et al. Evaluating late blight severity in potato crops using unmanned aerial vehicles and machine learning algorithms[J].Remote Sensing,2018,10(10):1513.