TW202303462A - Fault diagnosis system and method for state of radar system of roadside units - Google Patents
Fault diagnosis system and method for state of radar system of roadside units Download PDFInfo
- Publication number
- TW202303462A TW202303462A TW110125204A TW110125204A TW202303462A TW 202303462 A TW202303462 A TW 202303462A TW 110125204 A TW110125204 A TW 110125204A TW 110125204 A TW110125204 A TW 110125204A TW 202303462 A TW202303462 A TW 202303462A
- Authority
- TW
- Taiwan
- Prior art keywords
- coordinate
- points
- feature information
- radar
- detection points
- Prior art date
Links
Images
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
Description
本發明是有關於一種系統及其方法,且特別是有關於一種路側單元之雷達系統之狀態之錯誤診斷系統及其錯誤診斷方法。The present invention relates to a system and its method, and in particular to an error diagnosis system and an error diagnosis method for the status of a radar system of a roadside unit.
隨著5G世代的到來,智慧交通開始被廣泛討論,監測用之路側單元也將大量佈建。現存路側單元之雷達系統大多依靠人力調整與設置。With the advent of the 5G generation, smart transportation has begun to be widely discussed, and roadside units for monitoring will also be deployed in large numbers. The radar systems of the existing roadside units mostly rely on manual adjustment and setting.
然而,對於智慧交通中大量佈建之雷達系統僅仰賴人力監測並不切實際。However, it is impractical to rely solely on human monitoring for a large number of radar systems deployed in smart transportation.
本發明提出一種路側單元之雷達系統之狀態之錯誤診斷系統及其錯誤診斷方法,改善先前技術的問題。The present invention proposes an error diagnosis system and error diagnosis method for the state of the radar system of the roadside unit, which improves the problems of the prior art.
在本發明的一實施例中,本發明所提出的路側單元之雷達系統之狀態之錯誤診斷系統包含儲存裝置以及處理器。儲存裝置儲存至少一指令,處理器電性連接儲存裝置。處理器用以存取並執行至少一指令以:將路側單元之雷達系統之雷達資訊進行特徵提取,擷取出特徵資訊;將特徵資訊與預設錯誤進行權重訓練,以得出權重特徵資訊;將權重特徵資訊進行分類與診斷,以判斷路側單元之雷達系統是否發生預設錯誤。In an embodiment of the present invention, the error diagnosis system for the state of the radar system of the roadside unit proposed by the present invention includes a storage device and a processor. The storage device stores at least one instruction, and the processor is electrically connected to the storage device. The processor is used to access and execute at least one instruction to: perform feature extraction on the radar information of the radar system of the roadside unit to extract feature information; perform weight training on the feature information and default errors to obtain weight feature information; The feature information is classified and diagnosed to determine whether there is a preset error in the radar system of the roadside unit.
在本發明的一實施例中,處理器用以存取並執行至少一指令以:對於路側單元之雷達系統之雷達資訊以雷達偵測點(detection points)與雷達追蹤點(tracking points)進行蒐集,透過特徵提取所擷取出的特徵資訊包含複數特徵,複數特徵包含:複數偵測點的數量、複數偵測點於 x 座標之最大值、複數偵測點於 y 座標之最大值、複數偵測點於 x 座標之平均值、複數偵測點於 y 座標之平均值、複數偵測點於 x 座標之最小值、複數偵測點於 y 座標之最小值、複數偵測點於 x 座標之標準差、複數偵測點於 y 座標之標準差、複數偵測點於 x 座標之二次動差值、複數偵測點於 y 座標之二次動差值、複數追蹤點的數量、複數追蹤點於 x座標之平均值、複數追蹤點於 x 座標之最大值、複數追蹤點於 x 座標之最小值、複數追蹤點於 x 座標之標準差、複數追蹤點於 x 座標之二次動差值、複數追蹤點於 y 座標之平均值、複數追蹤點於 y 座標之最大值、複數追蹤點於 y 座標之最小值、複數追蹤點於 y 座標之標準差、複數追蹤點於 y 座標之二次動差值以及複數追蹤點於多個時間點所提取之軌跡特徵。In an embodiment of the present invention, the processor is used to access and execute at least one instruction: collect radar information of the radar system of the roadside unit using radar detection points and radar tracking points, The feature information extracted through feature extraction includes complex features. The complex features include: the number of complex detection points, the maximum value of the complex detection points at the x coordinate, the maximum value of the complex detection points at the y coordinate, and the complex detection points The average value of the x-coordinate, the average value of the multiple detection points at the y-coordinate, the minimum value of the multiple detection points at the x-coordinate, the minimum value of the multiple detection points at the y-coordinate, the standard deviation of the multiple detection points at the x-coordinate , the standard deviation of the complex detection points at the y coordinate, the quadratic dynamic difference of the complex detection points at the x coordinate, the quadratic dynamic difference of the complex detection points at the y coordinate, the number of the complex tracking points, the complex tracking points at The average value of the x-coordinate, the maximum value of the complex tracking points at the x-coordinate, the minimum value of the complex tracking points at the x-coordinate, the standard deviation of the complex tracking points at the x-coordinate, the quadratic dynamic difference of the complex tracking points at the x-coordinate, complex The average value of the tracking points at the y coordinate, the maximum value of the multiple tracking points at the y coordinate, the minimum value of the multiple tracking points at the y coordinate, the standard deviation of the multiple tracking points at the y coordinate, the quadratic dynamic difference of the multiple tracking points at the y coordinate value and track features extracted from multiple track points at multiple time points.
在本發明的一實施例中,預設錯誤為路側單元之雷達系統之一種或多種預設錯誤狀態,處理器用以存取並執行至少一指令以:執行權重訓練算法(attention),藉以將特徵資訊與一種或多種預設錯誤狀態進行權重訓練。In one embodiment of the present invention, the preset error is one or more preset error states of the radar system of the roadside unit, and the processor is used to access and execute at least one instruction to: execute a weight training algorithm (attention), so as to combine the features Information and one or more preset error states for weight training.
在本發明的一實施例中,權重訓練算法根據一種或多種預設錯誤狀態與特徵資訊的相關程度,進行權重訓練並給予特徵資訊中複數特徵不同的權重。In an embodiment of the present invention, the weight training algorithm performs weight training according to the degree of correlation between one or more preset error states and the feature information, and gives different weights to the multiple features in the feature information.
在本發明的一實施例中,處理器用以存取並執行至少一指令以:執行錯誤診斷神經網路(Diagnosis-Net),藉以將路側單元之雷達系統之特徵資訊或權重特徵資訊進行分類與診斷。In one embodiment of the present invention, the processor is used to access and execute at least one instruction to: execute a fault diagnosis neural network (Diagnosis-Net), so as to classify and compare the characteristic information or weight characteristic information of the radar system of the roadside unit diagnosis.
在本發明的一實施例中,本發明所提出的路側單元之雷達系統之狀態之錯誤診斷系統的錯誤診斷方法包含以下步驟:(A)將路側單元之雷達系統之雷達資訊進行特徵提取,擷取出特徵資訊;(B)將特徵資訊與預設錯誤進行權重訓練,以得出權重特徵資訊;(C)將權重特徵資訊進行分類與診斷,以判斷路側單元之雷達系統是否發生預設錯誤。In one embodiment of the present invention, the error diagnosis method of the error diagnosis system for the status of the radar system of the roadside unit proposed by the present invention includes the following steps: (A) performing feature extraction on the radar information of the radar system of the roadside unit, extracting Extract feature information; (B) perform weight training on feature information and default errors to obtain weight feature information; (C) classify and diagnose weight feature information to determine whether there is a default error in the radar system of the roadside unit.
在本發明的一實施例中,步驟(A)包含:對於路側單元之雷達系統之雷達資訊以雷達偵測點(detection points)與雷達追蹤點(tracking points)進行蒐集,透過特徵提取所擷取出的特徵資訊包含複數特徵,複數特徵包含:複數偵測點的數量、複數偵測點於 x 座標之最大值、複數偵測點於 y 座標之最大值、複數偵測點於 x 座標之平均值、複數偵測點於 y 座標之平均值、複數偵測點於 x 座標之最小值、複數偵測點於 y 座標之最小值、複數偵測點於 x 座標之標準差、複數偵測點於 y 座標之標準差、複數偵測點於 x 座標之二次動差值、複數偵測點於 y 座標之二次動差值、複數追蹤點的數量、複數追蹤點於 x座標之平均值、複數追蹤點於 x 座標之最大值、複數追蹤點於 x 座標之最小值、複數追蹤點於 x 座標之標準差、複數追蹤點於 x 座標之二次動差值、複數追蹤點於 y 座標之平均值、複數追蹤點於 y 座標之最大值、複數追蹤點於 y 座標之最小值、複數追蹤點於 y 座標之標準差、複數追蹤點於 y 座標之二次動差值以及複數追蹤點於多個時間點所提取之軌跡特徵。In one embodiment of the present invention, the step (A) includes: collecting the radar information of the radar system of the roadside unit using radar detection points and radar tracking points, and extracting them through feature extraction The feature information includes complex features, and the complex features include: the number of complex detection points, the maximum value of complex detection points at the x coordinate, the maximum value of the complex detection points at the y coordinate, and the average value of the complex detection points at the x coordinate , the average value of the complex detection points at the y-coordinate, the minimum value of the complex detection points at the x-coordinate, the minimum value of the complex detection points at the y-coordinate, the standard deviation of the complex detection points at the x-coordinate, the complex detection point at The standard deviation of the y-coordinate, the quadratic dynamic difference of the multiple detection points at the x-coordinate, the quadratic dynamic difference of the multiple detection points at the y-coordinate, the number of multiple tracking points, the average value of the multiple tracking points at the x-coordinate, The maximum value of the complex tracking points at the x coordinate, the minimum value of the complex tracking points at the x coordinate, the standard deviation of the complex tracking points at the x coordinate, the quadratic dynamic difference of the complex tracking points at the x coordinate, the The average value, the maximum value of the multiple tracking points at the y coordinate, the minimum value of the multiple tracking points at the y coordinate, the standard deviation of the multiple tracking points at the y coordinate, the quadratic dynamic difference of the multiple tracking points at the y coordinate, and the multiple tracking points at Trajectory features extracted at multiple time points.
在本發明的一實施例中,預設錯誤為路側單元之雷達系統之一種或多種預設錯誤狀態,步驟(B)包含:執行權重訓練算法,藉以將特徵資訊與一種或多種預設錯誤狀態進行權重訓練。In one embodiment of the present invention, the default error is one or more default error states of the radar system of the roadside unit, and step (B) includes: executing a weight training algorithm to combine feature information with one or more default error states Do weight training.
在本發明的一實施例中,權重訓練算法根據一種或多種預設錯誤狀態與特徵資訊的相關程度,進行權重訓練並給予特徵資訊中複數特徵不同的權重。In an embodiment of the present invention, the weight training algorithm performs weight training according to the degree of correlation between one or more preset error states and the feature information, and gives different weights to the multiple features in the feature information.
在本發明的一實施例中,步驟(C)包含:執行錯誤診斷神經網路,藉以將路側單元之雷達系統之特徵資訊或權重特徵資訊進行分類與診斷。In one embodiment of the present invention, the step (C) includes: executing the error diagnosis neural network, so as to classify and diagnose the feature information or weighted feature information of the radar system of the roadside unit.
綜上所述,本發明之技術方案與現有技術相比具有明顯的優點和有益效果。藉由本發明的技術方案,自動化錯誤診斷,減少運營上之人力開銷,提高診斷效率。To sum up, the technical solution of the present invention has obvious advantages and beneficial effects compared with the prior art. With the technical solution of the present invention, the error diagnosis is automated, the labor cost in operation is reduced, and the diagnosis efficiency is improved.
以下將以實施方式對上述之說明作詳細的描述,並對本發明之技術方案提供更進一步的解釋。The above-mentioned description will be described in detail in the following implementation manners, and further explanations will be provided for the technical solution of the present invention.
為了使本發明之敘述更加詳盡與完備,可參照所附之圖式及以下所述各種實施例,圖式中相同之號碼代表相同或相似之元件。另一方面,眾所週知的元件與步驟並未描述於實施例中,以避免對本發明造成不必要的限制。In order to make the description of the present invention more detailed and complete, reference may be made to the accompanying drawings and various embodiments described below, and the same numbers in the drawings represent the same or similar elements. On the other hand, well-known elements and steps have not been described in the embodiments in order to avoid unnecessarily limiting the invention.
請參照第1圖,本發明之技術態樣是一種路側單元之雷達系統190之狀態之錯誤診斷系統100,其可應用在智慧交通,或是廣泛地運用在相關之技術環節。本技術態樣之路側單元之雷達系統之狀態之錯誤診斷系統100可達到相當的技術進步,並具有産業上的廣泛利用價值。以下將搭配第1圖來說明路側單元之雷達系統之狀態之錯誤診斷系統100之具體實施方式。Please refer to FIG. 1 , the technical aspect of the present invention is a state
應瞭解到,路側單元之雷達系統190之狀態之錯誤診斷系統100的多種實施方式搭配第1圖進行描述。於以下描述中,為了便於解釋,進一步設定許多特定細節以提供一或多個實施方式的全面性闡述。然而,本技術可在沒有這些特定細節的情況下實施。於其他舉例中,為了有效描述這些實施方式,已知結構與裝置以方塊圖形式顯示。此處使用的「舉例而言」的用語,以表示「作為例子、實例或例證」的意思。此處描述的作為「舉例而言」的任何實施例,無須解讀為較佳或優於其他實施例。It should be understood that various implementations of the
第1圖是依照本發明一實施例之一種路側單元之雷達系統190之狀態之錯誤診斷系統100的方塊圖。如第1圖所示,路側單元之雷達系統之狀態之錯誤診斷系統100包含儲存裝置110、處理器120以及顯示器130。舉例而言,儲存裝置110可為硬碟、快閃儲存裝置或其他儲存媒介,處理器120可為中央處理器,顯示器130可為內建顯示器或外接螢幕。FIG. 1 is a block diagram of a state
在架構上,錯誤診斷系統100電性連接路側單元之雷達系統190,儲存裝置110電性連接處理器120,處理器120電性連接顯示器130。實作上,舉例而言,路側單元之雷達系統190可包含多個雷達以分配給各個路側單元。應瞭解到,於實施方式與申請專利範圍中,涉及『電性連接』之描述,其可泛指一元件透過其他元件而間接電氣耦合至另一元件,或是一元件無須透過其他元件而直接電連結至另一元件。舉例而言,儲存裝置110可為內建儲存裝置直接電連結至處理器120,或是儲存裝置110可為外部儲存設備透過線路間接連線至處理器120。In terms of architecture, the
於使用時,儲存裝置110儲存至少一指令,處理器120用以存取並執行至少一指令以:將路側單元之雷達系統190之資訊進行特徵提取,擷取出特徵資訊;將特徵資訊與預設錯誤進行權重訓練,以得出權重特徵資訊;將權重特徵資訊進行分類與診斷,以判斷路側單元之雷達系統是否發生預設錯誤。顯示器130可顯示相關的診斷結果。When in use, the
具體而言,本發明揭露之錯誤診斷系統100的核心演算法包含三部分:將路側單元之雷達系統190之資訊進行特徵提取、將提取之特徵資訊與相應錯誤之權重訓練、狀態分類訓練及評斷指標設計。在實施演算法前,首先針對路側單元之雷達系統190所產生之狀態蒐集相關雷達資訊,並透過特徵提取取得重要之雷達狀態特徵,為了加速錯誤診斷模型收斂速度,先將提取之數據進行正規化預處理;接著,針對預處理後的資料進行分類與訓練,最後,針對分類之結果進行判斷所屬之路側單元之雷達系統狀態。Specifically, the core algorithm of the
實作上,舉例而言,路側單元之雷達系統190以型號IWR6843為例進行說明,但不限制本發明。本例假設路側單元之雷達系統190可能有多種設置狀態,其包括:正常(Normality)、路側單元之雷達系統190之過度下頃(excessive down tilt,ED)、路側單元之雷達系統190之過度上頃(excessive up tilt,EU)、路側單元之雷達系統190之水平角順時鐘旋轉(azimuth clockwise rotates,ACR)、路側單元之雷達系統190之水平角逆時鐘旋轉(azimuth counter-clockwise rotates,ACCR)及路側單元之雷達系統190之偏移(shift)等狀態,此發明的適用範圍包含但不限於前述狀態。另外,本例之路側單元之雷達系統190之資訊以雷達偵測點(detection points)與雷達追蹤點(tracking points)進行蒐集,透過特徵提取之技術擷取重要之雷達系統資訊,再將經由特徵提取後之資料饋入錯誤診斷模型進行錯誤診斷,藉此判斷雷達系統190之狀態。In practice, for example, the radar system 190 of the roadside unit is described by taking the model IWR6843 as an example, but it does not limit the present invention. In this example, it is assumed that the radar system 190 of the RSU may have multiple configuration states, including: Normality, excessive down tilt (ED) of the radar system 190 of the RSU, and excessive down tilt of the RSU radar system 190 Excessive up tilt (EU), the horizontal angle of the radar system 190 of the roadside unit rotates clockwise (azimuth clockwise rotates, ACR), the horizontal angle of the radar system 190 of the roadside unit rotates counterclockwise (azimuth counter-clockwise rotates, ACCR) and the state of shift (shift) of the radar system 190 of the roadside unit, the scope of application of this invention includes but is not limited to the aforementioned state. In addition, in this example, the information of the radar system 190 of the roadside unit is collected by means of radar detection points and radar tracking points, and important radar system information is extracted through feature extraction technology, and then passed through the features The extracted data is fed into the error diagnosis model for error diagnosis, thereby judging the state of the radar system 190 .
在本發明的一實施例中,路側單元之雷達系統190之狀態之錯誤診斷系統100之核心演算法乃採用機器學習技術完成,其大致分為雷達資訊特徵提取、權重訓練和錯誤診斷三個步部分。In one embodiment of the present invention, the core algorithm of the
關於雷達資訊特徵提取,在本發明的一實施例中,處理器120用以存取並執行至少一指令以:對於路側單元之雷達系統之雷達資訊以雷達偵測點(detection points)與雷達追蹤點(tracking points)進行蒐集,透過特徵提取所擷取出的特徵資訊包含複數特徵,複數特徵包含:複數偵測點的數量、複數偵測點於 x 座標之最大值、複數偵測點於 y 座標之最大值、複數偵測點於 x 座標之平均值、複數偵測點於 y 座標之平均值、複數偵測點於 x 座標之最小值、複數偵測點於 y 座標之最小值、複數偵測點於 x 座標之標準差、複數偵測點於 y 座標之標準差、複數偵測點於 x 座標之二次動差值、複數偵測點於 y 座標之二次動差值、複數追蹤點的數量、複數追蹤點於 x座標之平均值、複數追蹤點於 x 座標之最大值、複數追蹤點於 x 座標之最小值、複數追蹤點於 x 座標之標準差、複數追蹤點於 x 座標之二次動差值、複數追蹤點於 y 座標之平均值、複數追蹤點於 y 座標之最大值、複數追蹤點於 y 座標之最小值、複數追蹤點於 y 座標之標準差、複數追蹤點於 y 座標之二次動差值、複數追蹤點於多個時間點所提取之軌跡特徵與/或其他特徵。With regard to radar information feature extraction, in an embodiment of the present invention, the
接下來,關於權重訓練,在本發明的一實施例中,上述預設錯誤為路側單元之雷達系統之一種或多種預設錯誤狀態(如:過度下頃、過度上頃、水平角順時鐘旋轉、水平角逆時鐘旋轉、偏移…等),處理器120用以存取並執行至少一指令以:執行權重訓練算法(attention),藉以將特徵資訊與一種或多種預設錯誤狀態進行權重訓練。權重訓練算法根據一種或多種預設錯誤狀態與特徵資訊的相關程度,進行權重訓練並給予特徵資訊中複數特徵不同的權重。Next, with regard to weight training, in one embodiment of the present invention, the above-mentioned preset error is one or more preset error states of the radar system of the roadside unit (such as: excessive downward, excessive upward, horizontal angle clockwise rotation , horizontal angle counterclockwise rotation, offset, etc.), the
實作上,不同的雷達資訊對於不同的路側單元之雷達系統狀態錯誤症狀將有不同程度的相關性,例如:路側單元之雷達系統190過度下頃可能與追蹤點於y座標之平均值、追蹤點於y座標之最大值、追蹤點於y座標之最小值、追蹤點於y座標之標準差、追蹤點於多個時間點所提取之軌跡特徵相關性較高,與偵測點數量、偵測點於x座標之平均值、偵測點於x座標之標準差、追蹤點數量、追蹤點於x座標之平均值、追蹤點於x座標之最大值、追蹤點於x座標之最小值、追蹤點於x座標之標準差相關性較低,若單純使用人類的專業知識及相應路側單元之雷達資訊做判斷與分析,將可能因誤判相關性,而導致診斷精準度下降之結果。因此,為了能夠使錯誤診斷系統100能夠學習路側單元之雷達系統190之狀態及相應雷達資訊之間的相關性,在進行錯誤診斷前,執行權重訓練算法(attention),使錯誤診斷系統100根據狀態與路側單元之雷達系統之雷達資訊的相關程度,訓練並給予不同的權重,以提升診斷精準度。In practice, different radar information will have different degrees of correlation with the radar system status error symptoms of different roadside units, for example: the radar system 190 of the roadside unit is excessively down, which may be related to the average value of the tracking point at the y coordinate, the tracking The maximum value of the point at the y-coordinate, the minimum value of the tracking point at the y-coordinate, the standard deviation of the tracking point at the y-coordinate, and the track features extracted at multiple time points are highly correlated with the number of detection points, detection The average value of the measuring point at the x-coordinate, the standard deviation of the detection point at the x-coordinate, the number of tracking points, the average value of the tracking point at the x-coordinate, the maximum value of the tracking point at the x-coordinate, the minimum value of the tracking point at the x-coordinate, The correlation of the standard deviation of the tracking point at the x-coordinate is low. If only human professional knowledge and the radar information of the corresponding roadside unit are used for judgment and analysis, it may lead to a decrease in diagnostic accuracy due to misjudgment of the correlation. Therefore, in order to enable the
實作上,舉例而言,權重訓練算法(attention)為一深度學習之機制,其目標是讓模型對重要訊息重點關注並充分學習吸收,藉由給予相關性高之訊息較高的權重,強化其重要程度;給予相關性低之訊息較低或0的權重,降低其影響力,以達到其目標。舉例而言,權重訓練算法(attention)架構包含三個隱藏全連接層(hidden fully connected layer),分別由50、20與6個神經元(neuron)所組成,再以softmax為激勵函數處理非線性特性(non-linear property)。整體權重訓練首先將各狀態及其相應之雷達資訊饋入attention架構進行權重訓練,經由attention訓練出各狀態之權重配置後,將輸入之雷達資訊與相應之權重進行阿達瑪乘積(Hadamard product),省去權重矩陣之存取,以加快計算速度,即完成各狀態之權重訓練。In practice, for example, the weight training algorithm (attention) is a deep learning mechanism. Its goal is to let the model focus on important information and fully learn and absorb it. By giving higher weights to highly relevant information, it strengthens Its degree of importance; giving low or 0 weight to information with low correlation, reducing its influence, so as to achieve its goal. For example, the weight training algorithm (attention) architecture includes three hidden fully connected layers (hidden fully connected layers), respectively composed of 50, 20 and 6 neurons (neurons), and then uses softmax as the activation function to handle nonlinearity Features (non-linear property). The overall weight training first feeds each state and its corresponding radar information into the attention framework for weight training. After the weight configuration of each state is trained through attention, the Hadamard product is performed on the input radar information and the corresponding weight. The access to the weight matrix is omitted to speed up the calculation, that is, the weight training of each state is completed.
關於分類與診斷,實作上,舉例而言,為了達到同時偵測路側單元之雷達系統190之多種錯誤發生之情況,假設一共有N類標籤,本發明採用以sigmoid為激活函數(activation function)之N維向量做為分類器輸出,當預測值大於0.5則判斷有該類錯誤發生。sigmoid函數的值域為[0,1],可將其看作是分類器認為有某錯誤發生的機率,判斷目前路側單元之雷達系統發生何種錯誤。分類之損失函數可為二元交叉熵(binary cross entropy, BCE)。Regarding classification and diagnosis, in practice, for example, in order to simultaneously detect multiple errors of the radar system 190 of the roadside unit, assuming that there are N types of tags, the present invention uses sigmoid as the activation function (activation function) The N-dimensional vector is used as the output of the classifier. When the predicted value is greater than 0.5, it is judged that there is such an error. The value range of the sigmoid function is [0,1], which can be regarded as the probability that the classifier believes that a certain error occurs, and judges what kind of error occurs in the radar system of the current roadside unit. The loss function for classification can be binary cross entropy (BCE).
在本發明的一實施例中,處理器120用以存取並執行至少一指令以:執行錯誤診斷神經網路(Diagnosis-Net),藉以將路側單元之雷達系統之特徵資訊或權重特徵資訊進行分類與診斷。In an embodiment of the present invention, the
實作上,舉例而言,錯誤診斷神經網路(Diagnosis-Net)架構包含四個隱藏全連接層(hidden fully connected layer),分別由30、20、10與5個神經元所組成,其中每一連接層之所有神經元都與下一層之所有神經元相連,而設計每一層之神經元數量遞減希望能萃取保留相對重要的特徵,幫助機器學習。透過多層數訓練學習後,可以判斷出輸入之路側單元之雷達系統狀態診斷資訊相對應之錯誤因素為何,然而,前三層搭配整流線性單元(rectified linear unit, ReLU)作為激活函數,最後一層搭配sigmoid作為激活函數,主要原因為Sigmoid會輸出一介於0-1之間的值,代表著此輸入資料對於此診斷因素結果之信心程度。In practice, for example, the Diagnosis-Net architecture includes four hidden fully connected layers, consisting of 30, 20, 10 and 5 neurons, each of which All neurons in a connection layer are connected to all neurons in the next layer, and the number of neurons in each layer is designed to decrease in order to extract and retain relatively important features to help machine learning. After multi-layer training and learning, it is possible to determine the error factors corresponding to the radar system status diagnostic information input to the roadside unit. However, the first three layers use a rectified linear unit (ReLU) as the activation function, and the last layer With sigmoid as the activation function, the main reason is that Sigmoid will output a value between 0-1, which represents the confidence level of the input data for the result of this diagnostic factor.
承上,依實務選定臨界值(threshold)將sigmoid輸出結果二分類為0或1,分別代表無或有偵測出此路側單元之雷達系統狀態錯誤。將此錯誤診斷神經網路(Diagnosis-Net)架構結合第一部分所提及之權重訓練算法(attention)與最終之臨界值設計則為此路側單元之雷達系統190自我狀態診斷專利之完整架構。實作上,能夠針對多重錯誤進行分類,機器學習能透過Sigmoid分辨輸入資料相對應錯誤診斷因素,設計不同臨界值而予以分類。Continuing from the above, the threshold value (threshold) is selected according to the practice to classify the sigmoid output results as 0 or 1, which respectively represent no or detected errors in the radar system status of the roadside unit. Combining this error diagnosis neural network (Diagnosis-Net) architecture with the weight training algorithm (attention) mentioned in the first part and the final threshold design is the complete architecture of the 190 self-state diagnosis patent for the radar system of this roadside unit. In practice, it is possible to classify multiple errors. Machine learning can identify the error diagnosis factors corresponding to the input data through Sigmoid, and design different thresholds for classification.
為了對上述路側單元之雷達系統之狀態之錯誤診斷系統100的錯誤診斷方法做更進一步的闡述,請同時參照第1~2圖,第2圖是依照本發明一實施例之一種路側單元之雷達系統190之狀態之錯誤診斷系統100的錯誤診斷方法200的流程圖。如第2圖所示,錯誤診斷方法200包含步驟S201~S203(應瞭解到,在本實施例中所提及的步驟,除特別敘明其順序者外,均可依實際需要調整其前後順序,甚至可同時或部分同時執行)。In order to further elaborate the error diagnosis method of the
錯誤診斷方法200可以採用非暫態電腦可讀取記錄媒體上的電腦程式產品的形式,此電腦可讀取記錄媒體具有包含在介質中的電腦可讀取的複數個指令。適合的記錄媒體可以包括以下任一者:非揮發性記憶體,例如:唯讀記憶體(ROM)、可程式唯讀記憶體(PROM)、可抹拭可程式唯讀記憶體(EPROM)、電子抹除式可程式唯讀記憶體(EEPROM);揮發性記憶體,例如:靜態存取記憶體(SRAM)、動態存取記憶體(SRAM)、雙倍資料率隨機存取記憶體(DDR-RAM);光學儲存裝置,例如:唯讀光碟(CD-ROM)、唯讀數位多功能影音光碟(DVD-ROM);磁性儲存裝置,例如:硬碟機、軟碟機。
於步驟S201,將路側單元之雷達系統190之雷達資訊進行特徵提取,擷取出特徵資訊。於步驟S202,將特徵資訊與預設錯誤進行權重訓練,以得出權重特徵資訊。於步驟S203,將權重特徵資訊進行分類與診斷,以判斷路側單元之雷達系統是否發生預設錯誤。In step S201, feature extraction is performed on the radar information of the radar system 190 of the roadside unit, and the feature information is extracted. In step S202, weight training is performed on the feature information and default errors to obtain weighted feature information. In step S203, the weighted feature information is classified and diagnosed to determine whether the radar system of the roadside unit has a preset error.
在本發明的一實施例中,於步驟S201,對於路側單元之雷達系統之雷達資訊以雷達偵測點(detection points)與雷達追蹤點(tracking points)進行蒐集,透過特徵提取所擷取出的特徵資訊包含複數特徵,複數特徵包含:複數偵測點的數量、複數偵測點於 x 座標之最大值、複數偵測點於 y 座標之最大值、複數偵測點於 x 座標之平均值、複數偵測點於 y 座標之平均值、複數偵測點於 x 座標之最小值、複數偵測點於 y 座標之最小值、複數偵測點於 x 座標之標準差、複數偵測點於 y 座標之標準差、複數偵測點於 x 座標之二次動差值、複數偵測點於 y 座標之二次動差值、複數追蹤點的數量、複數追蹤點於 x座標之平均值、複數追蹤點於 x 座標之最大值、複數追蹤點於 x 座標之最小值、複數追蹤點於 x 座標之標準差、複數追蹤點於 x 座標之二次動差值、複數追蹤點於 y 座標之平均值、複數追蹤點於 y 座標之最大值、複數追蹤點於 y 座標之最小值、複數追蹤點於 y 座標之標準差、複數追蹤點於 y 座標之二次動差值、複數追蹤點於多個時間點所提取之軌跡特徵。In one embodiment of the present invention, in step S201, the radar information of the radar system of the roadside unit is collected using radar detection points and radar tracking points, and features extracted through feature extraction The information includes complex features, which include: the number of complex detection points, the maximum value of complex detection points at the x coordinate, the maximum value of the complex detection points at the y coordinate, the average value of the complex detection points at the x coordinate, the complex The average value of the detection points at the y coordinate, the minimum value of the multiple detection points at the x coordinate, the minimum value of the multiple detection points at the y coordinate, the standard deviation of the multiple detection points at the x coordinate, and the y coordinate of the multiple detection points The standard deviation, the quadratic dynamic difference of multiple detection points at the x-coordinate, the quadratic dynamic difference of multiple detection points at the y-coordinate, the number of multiple tracking points, the average value of multiple tracking points at the x-coordinate, complex tracking The maximum value of the point at the x-coordinate, the minimum value of the multiple tracking points at the x-coordinate, the standard deviation of the multiple tracking points at the x-coordinate, the quadratic dynamic difference of the multiple tracking points at the x-coordinate, the average value of the multiple tracking points at the y-coordinate , the maximum value of the complex tracking points at the y coordinate, the minimum value of the complex tracking points at the y coordinate, the standard deviation of the complex tracking points at the y coordinate, the quadratic dynamic difference of the complex tracking points at the y coordinate, the multiple tracking points at multiple Trajectory features extracted at time points.
在本發明的一實施例中,預設錯誤為路側單元之雷達系統190之一種或多種預設錯誤狀態,於步驟S202,執行權重訓練算法,藉以將特徵資訊與一種或多種預設錯誤狀態進行權重訓練。In one embodiment of the present invention, the default error is one or more default error states of the radar system 190 of the roadside unit. In step S202, a weight training algorithm is executed to combine feature information with one or more default error states Weight training.
在本發明的一實施例中,權重訓練算法根據一種或多種預設錯誤狀態與特徵資訊的相關程度,進行權重訓練並給予特徵資訊中複數特徵不同的權重。In an embodiment of the present invention, the weight training algorithm performs weight training according to the degree of correlation between one or more preset error states and the feature information, and gives different weights to the multiple features in the feature information.
在本發明的一實施例中,於步驟S203,執行錯誤診斷神經網路,藉以將路側單元之雷達系統之特徵資訊或權重特徵資訊進行分類與診斷。In one embodiment of the present invention, in step S203 , an error diagnosis neural network is executed to classify and diagnose the characteristic information or weighted characteristic information of the radar system of the roadside unit.
綜上所述,本發明之技術方案與現有技術相比具有明顯的優點和有益效果。藉由本發明的路側單元之雷達系統190之狀態之錯誤診斷系統100及其錯誤診斷方法200,自動化錯誤診斷,減少運營上之人力開銷,提高診斷效率。To sum up, the technical solution of the present invention has obvious advantages and beneficial effects compared with the prior art. With the
雖然本發明已以實施方式揭露如上,然其並非用以限定本發明,任何熟習此技藝者,在不脫離本發明之精神和範圍內,當可作各種之更動與潤飾,因此本發明之保護範圍當視後附之申請專利範圍所界定者為準。Although the present invention has been disclosed above in terms of implementation, it is not intended to limit the present invention. Anyone skilled in this art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention The scope shall be defined by the appended patent application scope.
為讓本發明之上述和其他目的、特徵、優點與實施例能更明顯易懂,所附符號之說明如下: 100:錯誤診斷系統 110:儲存裝置 120:處理器 130:顯示器 190:路側單元之雷達系統 200:錯誤診斷方法 S201~S203:步驟 In order to make the above and other objects, features, advantages and embodiments of the present invention more obvious and understandable, the accompanying symbols are explained as follows: 100: Error diagnosis system 110: storage device 120: Processor 130: Display 190:Radar system of roadside unit 200: Error diagnosis method S201~S203: Steps
為讓本發明之上述和其他目的、特徵、優點與實施例能更明顯易懂,所附圖式之說明如下: 第1圖是依照本發明一實施例之一種路側單元之雷達系統之狀態之錯誤診斷系統的方塊圖;以及 第2圖是依照本發明一實施例之一種路側單元之雷達系統之狀態之錯誤診斷系統的錯誤診斷方法的流程圖。 In order to make the above and other objects, features, advantages and embodiments of the present invention more clearly understood, the accompanying drawings are described as follows: FIG. 1 is a block diagram of an error diagnosis system for the status of a radar system of a roadside unit according to an embodiment of the present invention; and FIG. 2 is a flow chart of an error diagnosis method of an error diagnosis system for the status of a radar system of a roadside unit according to an embodiment of the present invention.
200:錯誤診斷方法 200: Error diagnosis method
S201~S203:步驟 S201~S203: steps
Claims (10)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110125204A TWI771098B (en) | 2021-07-08 | 2021-07-08 | Fault diagnosis system and method for state of radar system of roadside units |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
TW110125204A TWI771098B (en) | 2021-07-08 | 2021-07-08 | Fault diagnosis system and method for state of radar system of roadside units |
Publications (2)
Publication Number | Publication Date |
---|---|
TWI771098B TWI771098B (en) | 2022-07-11 |
TW202303462A true TW202303462A (en) | 2023-01-16 |
Family
ID=83439447
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
TW110125204A TWI771098B (en) | 2021-07-08 | 2021-07-08 | Fault diagnosis system and method for state of radar system of roadside units |
Country Status (1)
Country | Link |
---|---|
TW (1) | TWI771098B (en) |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109325525A (en) * | 2018-08-31 | 2019-02-12 | 阿里巴巴集团控股有限公司 | Sample attribute assessment models training method, device and server |
TWI717655B (en) * | 2018-11-09 | 2021-02-01 | 財團法人資訊工業策進會 | Feature determination apparatus and method adapted to multiple object sizes |
CN112488142A (en) * | 2019-09-12 | 2021-03-12 | 中移(苏州)软件技术有限公司 | Radar fault prediction method and device and storage medium |
CN110736968B (en) * | 2019-10-16 | 2021-10-08 | 清华大学 | Radar abnormal state diagnosis method based on deep learning |
CN113030881A (en) * | 2019-12-09 | 2021-06-25 | 上海禾赛科技股份有限公司 | Point cloud rationality diagnosis method for laser radar, and vehicle including the same |
CN112986950A (en) * | 2020-12-25 | 2021-06-18 | 南京理工大学 | Single-pulse laser radar echo feature extraction method based on deep learning |
CN113033786B (en) * | 2021-05-21 | 2021-08-13 | 北京航空航天大学 | Fault diagnosis model construction method and device based on time convolution network |
-
2021
- 2021-07-08 TW TW110125204A patent/TWI771098B/en active
Also Published As
Publication number | Publication date |
---|---|
TWI771098B (en) | 2022-07-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Islam et al. | Revisiting salient object detection: Simultaneous detection, ranking, and subitizing of multiple salient objects | |
CN107392015B (en) | A kind of intrusion detection method based on semi-supervised learning | |
CN102147851B (en) | Device and method for judging specific object in multi-angles | |
US8724904B2 (en) | Anomaly detection in images and videos | |
CN109087510A (en) | traffic monitoring method and device | |
US10235629B2 (en) | Sensor data confidence estimation based on statistical analysis | |
CN110399238A (en) | A kind of disk failure method for early warning, device, equipment and readable storage medium storing program for executing | |
CN106778908A (en) | A kind of novelty detection method and apparatus | |
CN108919059A (en) | A kind of electric network failure diagnosis method, apparatus, equipment and readable storage medium storing program for executing | |
CN109961030A (en) | Pavement patching information detecting method, device, equipment and storage medium | |
WO2023123685A1 (en) | Method and apparatus for evaluating service life of wind power generator set | |
CN108288020A (en) | Video shelter detecting system based on contextual information and method | |
CN109376736A (en) | A kind of small video target detection method based on depth convolutional neural networks | |
CN112541536A (en) | Under-sampling classification integration method, device and storage medium for credit scoring | |
CN111753877B (en) | Product quality detection method based on deep neural network migration learning | |
Wen et al. | A new method for identifying the ball screw degradation level based on the multiple classifier system | |
Seibold et al. | Self-guided multiple instance learning for weakly supervised thoracic diseaseclassification and localizationin chest radiographs | |
US11132790B2 (en) | Wafer map identification method and computer-readable recording medium | |
Liu et al. | Defect detection of the surface of wind turbine blades combining attention mechanism | |
CN103971362A (en) | Synthetic aperture radar (SAR) imagine change detection based on histogram and elite genetic clustering algorithm | |
TWI771098B (en) | Fault diagnosis system and method for state of radar system of roadside units | |
CN117195114A (en) | Chemical production line identification method and system | |
CN116907772A (en) | Self-diagnosis and fault source identification method and system of bridge structure monitoring sensor | |
CN107480647B (en) | Method for detecting abnormal behaviors in real time based on inductive consistency abnormality detection | |
CN115794451A (en) | Execution strategy prediction method, device and system based on health state of storage equipment |