一、介紹

鋼軌表面缺陷檢測是指通過使用各種技術(shù)手段和設備，對鋼軌表面進行檢查和測量，以確定是否存在裂紋、掉塊、剝離、銹蝕等缺陷的過程。這些缺陷可能會對鐵路運輸?shù)陌踩头€(wěn)定性產(chǎn)生影響，因此及時進行檢測和修復非常重要。鋼軌表面缺陷檢測通常采用無損檢測技術(shù)，如超聲檢測、渦流檢測等，以確保在不損害鋼軌的前提下進行準確的檢測。

二、數(shù)據(jù)

鋼軌表面缺陷數(shù)據(jù)通常包括缺陷的類型、位置、尺寸以及嚴重程度等信息。這些數(shù)據(jù)可以通過各種檢測設備和技術(shù)獲取，如激光掃描儀、高清相機等。這些數(shù)據(jù)對于評估鋼軌的狀態(tài)、制定維護計劃以及確保鐵路運輸?shù)陌踩哂兄匾饬x。通過對這些數(shù)據(jù)的分析和處理，可以實現(xiàn)對鋼軌表面缺陷的準確檢測和分類，有助于提高鋼軌維護的效率和安全性。

三、獲取

本數(shù)據(jù)集原始是一個4類的圖像分類數(shù)據(jù)集，總共有4個類別（通過標注處理，成為目標檢測數(shù)據(jù)集，含xml標簽文件，聯(lián)系小編獲取）：

根據(jù)缺陷類別，進行標注：

得到2個類別的缺陷數(shù)據(jù)集，可用于目標檢測任務，適用于yolov3、yolov4、yolov5、yolov6、yolov7、yolov8等算法模型訓練任務。

目前鋼軌表面缺陷檢測存在的問題有：智能化程度低、鋼軌缺陷檢測研究較少、鋼軌表面材質(zhì)特殊，處理難度大。我們通過實地參觀考察發(fā)現(xiàn)，現(xiàn)有的大型鋼鐵軌梁廠如攀鋼、包鋼等仍采用人工目測法對鋼軌表面質(zhì)量進行監(jiān)控，生產(chǎn)效率低，對后續(xù)的工藝改進參考價值不大。通過調(diào)研國內(nèi)外文獻可知，目前比較成熟的鋼類產(chǎn)品缺陷檢測技術(shù)主要集中于鋼板，對冷態(tài)鋼軌的研究甚少。鋼軌是一種高反光性材質(zhì)，其表面灰度變化不大，因此鋼軌缺陷檢測對成像質(zhì)量以及缺陷分割算法有更高的要求。

通過讀取xml標簽文件，可以獲得類別名稱和標簽數(shù)量：

import os
import xml.etree.ElementTree as ET
import globdef count_type_num(indir):# 提取xml文件列表os.chdir(indir)annotations = os.listdir('.')annotations = glob.glob(str(annotations) + '*.xml')dict = {}  # 新建字典，用于存放各類標簽名及其對應的數(shù)目for i, file in enumerate(annotations):  # 遍歷xml文件# actual parsingin_file = open(file, encoding='utf-8')tree = ET.parse(in_file)root = tree.getroot()# 遍歷文件的所有標簽for obj in root.iter('object'):name = obj.find('name').textif (name in dict.keys()):dict[name] += 1  # 如果標簽不是第一次出現(xiàn)，則+1else:dict[name] = 1  # 如果標簽是第一次出現(xiàn)，則將該標簽名對應的value初始化為1# 打印結(jié)果print("各類標簽的數(shù)量分別為：")for key in dict.keys():print(key + ': ' + str(dict[key]))

四、最后

鋼軌是鐵路軌道的主要部件，起引導列車運行和直接承受車輛載荷的重要作用。隨著我國既有線路改造以及高速鐵路的快速發(fā)展，列車對鋼軌的運行壓力以及沖擊載荷越來越強，鋼軌表面產(chǎn)生的缺陷概率也越來越大。因此，采集鋼軌表面缺陷數(shù)據(jù)，并基于先進的算法進行檢測，是保障鐵路安全和穩(wěn)定運行的重要手段，具有極其重要的意義。

早期鋼軌缺陷檢測的主要手段是人工物探，該方法不僅效率低下，且無法形成客觀統(tǒng)一的檢測標準，正逐漸被其他方法所取代.隨后，超聲波、射線、滲透、渦流等鋼軌無損探傷技術(shù)的應用推動了檢測精度和檢測速度的相對提高，這些檢測方法雖然穿透能力強、操作安全，但容易受到外部干擾影響，檢測結(jié)果抽象且難以處理?；跈C器視覺的鋼軌缺陷檢測方法通過先進的視覺設備采集鋼軌表面圖像，根據(jù)算法對圖像進行處理，具有實時性、非接觸式等特點，能夠很好地運用于鋼軌缺陷檢測領(lǐng)域。閔永智等提出了將平滑濾波器與閾值分割相結(jié)合的鋼軌表面缺陷檢測方法，減輕了光照變化、軌面不平對檢測結(jié)果的影響，但該方法對背景圖像的自適應平滑過程運算量過大，實時性不強。Shi等針對光照及環(huán)境變化造成鋼軌圖像降質(zhì)的問題，提出了一種基于邊緣檢測算子改進的鋼軌缺陷檢測算法，改進后的算法可獲得具有完整邊緣信息的缺陷輪廓定位，但對復雜鋼軌圖像的檢測準確率較低。Tastimur等提出了一種基于形態(tài)學特征提取的鐵路缺陷檢測算法，利用霍夫變換和圖像處理技術(shù)對實時攝像機獲取的鋼軌圖像進行檢測，并通過形態(tài)學操作提取采集到的鋼軌圖像特征，實現(xiàn)對缺陷的識別，但復雜的圖像預處理過程容易受到光照不均等外部因素的影響，造成一定程度的漏檢.上述研究將傳統(tǒng)圖像處理技術(shù)與機器學習的方法相結(jié)合，設計了適用于特定場景下的鋼軌缺陷檢測方法，該類方法的檢測性能易受外部環(huán)境的影響，檢測速度難以滿足實時檢測要求。

Steel rails are the main components of railway tracks, playing an important role in guiding train operation and directly bearing vehicle loads. With the renovation of existing railway lines and the rapid development of high-speed railways in China, the operating pressure and impact load of trains on steel rails are becoming stronger, and the probability of defects on the surface of steel rails is also increasing. Therefore, collecting data on rail surface defects and detecting them based on advanced algorithms is an important means to ensure the safety and stable operation of railways, and has extremely important significance.The main method of early rail defect detection was manual geophysical exploration, which was not only inefficient but also unable to form objective and unified detection standards. It was gradually replaced by other methods. Subsequently, the application of non-destructive testing technologies such as ultrasound, radiation, penetration, and eddy current for steel rails has promoted the relative improvement of detection accuracy and speed. Although these detection methods have strong penetration ability and safe operation, they are easily affected by external interference, The detection results are abstract and difficult to process. The machine vision based rail defect detection method collects rail surface images through advanced visual equipment and processes the images based on algorithms. It has real-time and non-contact characteristics and can be well applied in the field of rail defect detection. Min Yongzhi et al. proposed a rail surface defect detection method that combines smooth filters with threshold segmentation, reducing the impact of lighting changes and uneven rail surface on the detection results. However, this method requires too much computation for the adaptive smoothing process of background images and lacks real-time performance. Shi et al. proposed an improved rail defect detection algorithm based on edge detection operator to address the issue of degraded rail images caused by lighting and environmental changes. The improved algorithm can obtain defect contour localization with complete edge information, but the detection accuracy for complex rail images is low. Tastimur et al. proposed a railway defect detection algorithm based on morphological feature extraction, which utilizes Hough transform and image processing technology to detect real-time camera captured rail images, and extracts collected rail image features through morphological operations to achieve defect recognition. However, complex image preprocessing processes are easily affected by external factors such as uneven lighting, Causing a certain degree of missed detection. The above research combines traditional image processing techniques with machine learning methods to design rail defect detection methods suitable for specific scenarios. The detection performance of these methods is easily affected by external environments, and the detection speed is difficult to meet real-time detection requirements.