Ryanflash/herislab-ca-training-data
收藏资源简介:
--- language: - en license: mit task_categories: - image-classification tags: - thermal-imaging - anomaly-detection - autoencoder - electrical-fault-detection - infrared size_categories: - 10K<n<100K --- # CA_Training_Data -- Convolutional Autoencoder (Track A) Curated dataset for training and evaluating the Convolutional Autoencoder anomaly detection model. ## Approach The autoencoder is trained **only on normal (no-fault) images**. At inference, high reconstruction error indicates an anomaly/fault. ## Structure ``` train/normal/ -- Normal images for autoencoder training electric_motor/ -- 168 PNG (Electric Motor Thermal Fault Diagnosis, no_fault class) induction_motor/ -- 20 BMP (Thermal Images of Induction Motor, Noload class) pv_om_inspection/ -- 7,836 TIFF (PV System O&M Inspection, double-row + single-row) pv_thermal_inspection/ -- 1,075 TIFF (PV System Thermal Inspection) solar_modules/ -- 2,302 JPG (Infrared Solar Modules, No-Anomaly class) test/normal/ -- Held-out normal images for threshold calibration electric_motor/ -- 28 PNG induction_motor/ -- 5 BMP test/fault/ -- Fault images for evaluating anomaly detection electric_motor/ -- 173 PNG (Electric Motor Thermal Fault Diagnosis, fault class) induction_motor/ -- 344 BMP (Thermal Images of Induction Motor, 10 fault conditions) ``` ## Total Counts | Split | Normal | Fault | Total | |-------|--------|-------|-------| | Train | 11,401 | 0 | 11,401 | | Test | 33 | 517 | 550 | ## Source Datasets | Directory | Source Dataset | Domain | |-----------|--------------|--------| | electric_motor | Electric Motor Thermal image Fault Diagnosis DATASET | Electrical (primary) | | induction_motor | Thermal Images of Induction Motor Dataset | Electrical (primary) | | pv_om_inspection | Photovoltaic System O&M inspection | Solar PV (adjacent) | | pv_thermal_inspection | Photovoltaic system thermal inspection | Solar PV (adjacent) | | solar_modules | Infrared Solar Modules (No-Anomaly only) | Solar PV (adjacent) | ## Notes - PV O&M files are prefixed `dr_` (double-row) and `sr_` (single-row) to avoid filename collisions - Solar module images were filtered from module_metadata.json (anomaly_class == "No-Anomaly") - Test/normal hold-out is ~14-20% of electrical equipment normal images - Image formats are mixed (PNG, BMP, TIFF, JPG) -- preprocessing/normalization is required before training
语言: - en 许可证:MIT 任务类别: - 图像分类 标签: - 热成像 - 异常检测 - 自编码器(autoencoder) - 电气故障检测 - 红外 样本规模:10000 < 样本量 < 100000 --- # CA_Training_Data —— 卷积自编码器(Convolutional Autoencoder)赛道A 本数据集为精选数据集,用于训练与评估卷积自编码器异常检测模型。 ## 方法 仅在正常(无故障)图像上训练自编码器。推理阶段,较高的重建误差即可判定存在异常/故障。 ## 数据集结构 train/normal/ -- 用于自编码器训练的正常图像 electric_motor/ -- 168张PNG图像(电动机热故障诊断,无故障类别) induction_motor/ -- 20张BMP图像(感应电动机热图像,空载类别) pv_om_inspection/ -- 7836张TIFF图像(光伏系统运维巡检,包含双列与单列阵列) pv_thermal_inspection/ -- 1075张TIFF图像(光伏系统热巡检) solar_modules/ -- 2302张JPG图像(红外太阳能组件,无异常类别) test/normal/ -- 用于阈值校准的预留正常样本 electric_motor/ -- 28张PNG图像 induction_motor/ -- 5张BMP图像 test/fault/ -- 用于异常检测评估的故障样本 electric_motor/ -- 173张PNG图像(电动机热故障诊断,故障类别) induction_motor/ -- 344张BMP图像(感应电动机热图像,包含10种故障工况) ## 样本总量统计 | 拆分集 | 正常样本数 | 故障样本数 | 总样本数 | |-------|--------|-------|-------| | 训练集 | 11401 | 0 | 11401 | | 测试集 | 33 | 517 | 550 | ## 源数据集信息 | 子目录 | 源数据集名称 | 应用领域 | |-----------|--------------|--------| | electric_motor | 电动机热图像故障诊断数据集 | 电气(核心领域) | | induction_motor | 感应电动机热图像数据集 | 电气(核心领域) | | pv_om_inspection | 光伏系统运维巡检数据集 | 太阳能光伏(关联领域) | | pv_thermal_inspection | 光伏系统热巡检数据集 | 太阳能光伏(关联领域) | | solar_modules | 红外太阳能组件数据集(仅无异常样本) | 太阳能光伏(关联领域) | ## 备注 - 光伏运维文件的文件名前缀为`dr_`(双列阵列)与`sr_`(单列阵列),以避免文件名冲突 - 太阳能组件图像均从`module_metadata.json`中筛选得到(异常类别为"No-Anomaly",即无异常) - 测试集正常样本约占电气设备正常图像的14%~20% - 图像格式混杂(包含PNG、BMP、TIFF、JPG),训练前需完成预处理与归一化操作



