# Computer-Rookie1 **Repository Path**: iamher321/computer-rookie1 ## Basic Information - **Project Name**: Computer-Rookie1 - **Description**: C#命令行执行Python脚本 - **Primary Language**: C# - **License**: Not specified - **Default Branch**: master - **Homepage**: https://gitee.com/computer-rookie1/computer-rookie1 - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 2 - **Created**: 2022-06-22 - **Last Updated**: 2022-06-22 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # C#命令行执行Python脚本 ## 一、环境 ### 1. 软件环境 1. Anaconda 2. PyCharm 3. Visual Studio 2015 ### 2. Python环境 1. opencv_python ``` pip install -i https://pypi.tuna.tsinghua.edu.cn/simple opencv_python ``` 2. matplotlib ``` pip install -i https://pypi.tuna.tsinghua.edu.cn/simple matplotlib ``` ## 二、Python代码 ```python # encoding: utf-8 """ @Time:2022/3/21 16:01 @Author: shujin sun @Desc: 缺陷检测测试 """ import cv2 as cv import numpy as np import matplotlib.pyplot as plt import math import sys # 中文显示 plt.rcParams['font.sans-serif'] = ['SimHei'] # 显示中文 plt.rcParams['axes.unicode_minus'] = False # 显示坐标轴负号 class DetectDefect: def __init__(self): self.little_img_file = '' self.big_img_file = '' self.big_img = np.array(0) self.little_img = np.array(0) self.roi_img = np.array(0) # 缺陷检测二值化结果 self.bw_img = np.array(0) # big image 的大小 self.w = 0 self.h = 0 # 重点关注区域，即中间区域 self.roi_rect = [0, 0, 0, 0] self.roi_patchs = [] # 缺陷检测阈值 self.thresh_value = 3000 # 所有的匹配方法 self.match_methods = ['cv.TM_CCOEFF', # 相关系数匹配法 'cv.TM_CCOEFF_NORMED', # 归一化版本 'cv.TM_CCORR', # 相关匹配法 'cv.TM_CCORR_NORMED', 'cv.TM_SQDIFF', # 平方差匹配法 'cv.TM_SQDIFF_NORMED'] def set_little_image(self, _file): self.little_img_file = _file self.little_img = cv.imread(_file)# cv.IMREAD_GRAYSCALE def set_big_image(self, _file): self.big_img_file = _file self.big_img = cv.imread(_file) print(self.big_img.shape) self.h, self.w = self.big_img.shape[:2] print('big image shape: ', self.w, self.h) def set_thresh_value(self, thresh_value): """设置缺陷检测阈值门限""" self.thresh_value = thresh_value def conduct_edge(self, src): '''边缘检测''' # 计算像素中位数 gray_img = cv.cvtColor(src, cv.COLOR_BGR2GRAY) median_intensity = np.median(gray_img) lower_threshold = int(max(0, (1.0 - 0.33) * median_intensity)) upper_threshold = int(min(255, (1.0 + 0.33) * median_intensity)) canny_img = cv.Canny(gray_img, lower_threshold, upper_threshold) plt.imshow(canny_img, cmap='gray') plt.title('canny edge detect result') plt.xticks([]), plt.yticks([]) plt.show() def conduct_binary_adaptive(self, src): """二值化操作,自适应方法""" src = cv.cvtColor(src, cv.COLOR_BGR2GRAY) bw_img = cv.adaptiveThreshold(src, 255, cv.ADAPTIVE_THRESH_MEAN_C,#cv.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY, 5, 7) return bw_img def conduct_binary(self, src): """二值化操作""" src = cv.cvtColor(src, cv.COLOR_BGR2GRAY) mean_intensity = np.mean(src) median_intensity = np.median(src) # print('图像灰度均值、中值：', mean_intensity, median_intensity) _, bw_img = cv.threshold(src, mean_intensity, 255, cv.THRESH_BINARY_INV+cv.THRESH_OTSU) return bw_img def match_template(self, big, little): img = big.copy() # eval函数，将字符串作为脚本代码来执行 idx = 3 # 1 or 3方法效果最好 print('选择的匹配算法：', self.match_methods[idx]) method = eval(self.match_methods[idx]) res = cv.matchTemplate(img, little, method) min_val, max_val, min_loc, max_loc = cv.minMaxLoc(res) top_left = max_loc h, w = little.shape[:2] bottom_right = (top_left[0] + w, top_left[1] + h) return top_left, bottom_right def split_patches(self): """异常检测 1.两种choice匹配 choice1：选择所有匹配算法计算，通过比较评估值，计算最优匹配结果 choice2：仅挑选一种匹配算法计算，通过前期测试验证，事先固化最稳健的算法 另外，可考虑图像偏转一定角度（极小范围，可多个值，该角度值若能用算法估算更佳），旋转后再进行匹配、检测 2.上下扩充 在roi区域中匹配结束之后，上下进行扩充选择，进行缺陷区域检测，防止漏网之鱼 3.二次匹配、检测 在结果中按照固定尺寸分割小块，随机选择1块，or二值化求取面积最大快作为模板块（或者在little_img中截取） 4.通过模板块，在其他区域进行相关或者相减计算（考虑先进行二值化），根据一定阈值，给出是否缺陷判决 在roi块中匹配结束之后，再转换至整体图像中位置（视需求而定，忽视此操作也可） """ top_left, bottom_right = self.match_template(self.roi_img, self.little_img) print('roi中间结果：', top_left, bottom_right) # 通过上下平移获取所有可能的匹配结果 h = self.little_img.shape[0] # 小图高度 n_upper = -int(top_left[1] / h) n_bottom = int((self.roi_img.shape[0] - bottom_right[1]) / h) patch_arrays = [] # 保存切片在ROI中的位置 tmp_img = self.roi_img.copy() for i in range(n_upper, n_bottom + 1): # 上下拓展，包括自身 t = top_left[1] + i * h b = bottom_right[1] + i * h if i != 0: t = t + int(i / abs(i)) b = b + int(i / abs(i)) print('i={0},范围：{1}-{2}'.format(i, t, b)) # 截取图像切片 patch_arrays.append([t, b, top_left[0], bottom_right[0]]) src = tmp_img[t:b, top_left[0]:bottom_right[0]] # 保存为文件 # save_name = '{0}_slice_{1}.bmp'.format(big_img_file[:-4], i) # print(src.shape, save_name) # cv.imwrite(save_name, src) # 显示中间匹配结果 cv.rectangle(tmp_img, (top_left[0], t), (bottom_right[0], b), 255, 2) self.roi_patchs = np.array(patch_arrays) print('切片位置：', self.roi_patchs, self.roi_patchs.shape) plt.imshow(tmp_img, cmap='gray') plt.title('ROI匹配中间结果') plt.show() def set_roi_rect(self, ratio_left=0.27, ratio_right=0.82, ratio_upper=0.2, ratio_bottom=0.8): '''设置中间关注区域，左右上下位置的ratio比例''' self.roi_rect[0] = int(self.w * ratio_left) self.roi_rect[1] = int(self.w * ratio_right) self.roi_rect[2] = int(self.h * ratio_upper) self.roi_rect[3] = int(self.h * ratio_bottom) def show_roi_img(self): """显示中间roi图像""" print('roi:', self.roi_rect) self.roi_img = self.big_img[self.roi_rect[2]:self.roi_rect[3], self.roi_rect[0]:self.roi_rect[1],:] # 保存中间roi图像 save_name = '{0}_roi.bmp'.format(big_img_file[:-4]) cv.imwrite(save_name, self.roi_img) tmp_img = self.big_img.copy() top_left = (self.roi_rect[1], self.roi_rect[3]) bottom_right = (self.roi_rect[0], self.roi_rect[2]) cv.rectangle(tmp_img, top_left, bottom_right, 255, 2) plt.subplot(131), plt.title('Big image') plt.imshow(tmp_img) plt.subplot(132), plt.title('roi image') plt.imshow(self.roi_img, cmap='gray') # plt.xticks([]), plt.yticks([]) plt.subplot(133), plt.title('little image') plt.imshow(self.little_img, cmap='gray') plt.show() def show_defect_image(self): '''显示缺陷检测结果图像''' tmp_img = self.big_img.copy() top_left = (50, 50) bottom_right = (100, 100) cv.rectangle(tmp_img, top_left, bottom_right, (255, 0, 0), 4) plt.imshow(tmp_img) plt.title('缺陷检测结果') plt.show() def defect_analyze_patchs(self): '''对切片逐个进行二值化和连通分析，返回连通分析结果''' is_defected = False img_output = self.big_img.copy() n_patches = self.roi_patchs.shape[0] roi_img_output = self.roi_img.copy() for i in range(0, n_patches): pos = self.roi_patchs[i, :] # patch 在roi中的位置,按照左右上下顺序排列 path_in_roi = pos[2], pos[0] print('patch 在ROI中位置：', path_in_roi) patch = self.roi_img[pos[0]:pos[1], pos[2]:pos[3]] # 每个小方块的检测结果 result, width_square = self.square_detect_in_patch(patch) for j in range(0, result.shape[0]): if result[j, 0] == 1: is_defected = True topleft = int(j*width_square + result[j, 1]), int(result[j, 2]) bottomright = int(topleft[0] + result[j, 3]), int(result[j, 4]) print('patch 中位置：', topleft, bottomright) # 在切片上画图 # cv.rectangle(patch, topleft, bottomright, (255, 0, 0), 10) # 将位置转换至ROI图像上, x,y,width,height pos_in_patch = topleft[0], topleft[1], result[j, 3], result[j, 4] pos_in_roi = self.cvt_pos_to_roi(pos_in_patch, path_in_roi) # 在ROI上画图 topleft, bottomright = self.cvt_top_bottom(pos_in_roi[0], pos_in_roi[1], pos_in_roi[2], pos_in_roi[3]) print('ROI中位置：', topleft, bottomright) cv.rectangle(roi_img_output, topleft, bottomright, (255, 0, 0), 5) # 将位置转换至整个图像上 pos_in_img = self.cvt_pos_to_image(pos_in_patch, path_in_roi) topleft, bottomright = self.cvt_top_bottom(pos_in_img[0], pos_in_img[1], pos_in_img[2], pos_in_img[3]) cv.rectangle(img_output, topleft, bottomright, (255, 0, 0), 5) plt.subplot(121), plt.imshow(roi_img_output) plt.title('ROI区域') plt.subplot(122), plt.imshow(img_output) title = '有缺陷' if is_defected else '正常' plt.title(title) plt.show() def cvt_top_bottom(self, x, y, width, height): '''将x,y,w,h转换为矩阵的topleft和bottomright''' topleft = x, y bottomright = x + width, y + height return topleft, bottomright def cvt_pos_to_roi(self, pos_in_patch, patch_in_roi): '''将patch坐标转换至roi上''' pos_in_roi= [] x = pos_in_patch[0] + patch_in_roi[0] y = pos_in_patch[1] + patch_in_roi[1] width = pos_in_patch[2] height = pos_in_patch[3] pos_in_roi = int(x), int(y), int(width), int(height) return pos_in_roi def cvt_pos_to_image(self, pos_in_patch, patch_in_roi): '''将patch坐标转换至image上''' pos_in_image = [] x = pos_in_patch[0] + patch_in_roi[0] + self.roi_rect[0] y = pos_in_patch[1] + patch_in_roi[1] + self.roi_rect[2] width = pos_in_patch[2] height = pos_in_patch[3] pos_in_image = int(x), int(y), int(width), int(height) return pos_in_image def white_ratio_analyze(self, src): """白色比例分析""" w, h = src.shape # print("ratio分析：", w, h, src.sum(), src.max()) ratio = src.sum() / (w * h * 255) return ratio def connect_component(self, src_in): """连通区域分析""" src = src_in.copy() n, labels, stats, centroids = cv.connectedComponentsWithStats(src, connectivity=8) # print('----' * 20, "\n", stats) # 计算面积 同 长宽比值，或者面积最大值 # 此处直接使用面积最大值 # stats：x y width, height, area，行数为n,首行为背景 array_ratio = stats[1:, -1] # print('剔除背景之后：', array_ratio) idx = array_ratio.argmax() + 1 max_ratio = array_ratio.max() # print('像素面积：', stats[:, 4], sum(stats[:, 4])) # print('面积最大值：', max_ratio, idx) # 根据像素点计算实际面积 mask1 = labels == idx # print('实际像素个数:', sum(sum(mask1))) # 不同的连通域赋予不同的颜色 output = np.zeros((labels.shape[0], labels.shape[1], 3), np.uint8) for i in range(1, n): mask = labels == i output[:, :, 0][mask] = np.random.randint(0, 255) output[:, :, 1][mask] = np.random.randint(0, 255) output[:, :, 2][mask] = np.random.randint(0, 255) # 矩形区域 x, y, width, height, areas = stats[idx, :] topleft = x, y bottomright = x + width, y + height cv.rectangle(output, topleft, bottomright, (255, 255, 0), 2, 1, 0) # 中心点 center = int(centroids[idx, 0]), int(centroids[idx, 1]) cv.circle(output, center, 5, (255, 0, 0), 1) # cv.putText(output,"{0}".format(areas),center,cv.FONT_HERSHEY_SIMPLEX,0.5,(255,0,0),1) # for idx_i in range(0,n): # # 矩形区域 # x, y, width, height, areas = stats[idx_i, :] # print(idx_i, '循环:', areas) # topleft = x, y # bottomright = x + width, y + height # #cv.rectangle(output, topleft, bottomright, (255, 255, 0), 1, 1, 0) # # 中心点 # center = int(centroids[idx_i, 0]), int(centroids[idx_i, 1]) # cv.circle(output, center, 5, (255, 0, 0), 1) # # cv.putText(output,"{0}".format(areas),center,cv.FONT_HERSHEY_SIMPLEX,0.5,(255,0,0),1) return output, stats[idx, :] def square_detect_in_patch(self, patch): # ROI图像中共11块 N = 11 result = np.zeros((N, 6)) h, w = patch.shape[:2] one_w = int(w / N) for i in range(0, N): # print(f'----------------------分块序号：{i}---------------------') # 抽取块，然后计算相关系数 tmp_img = patch[:, i * one_w + 3:(i + 1) * one_w + 3] bw_img = self.conduct_binary(tmp_img) # 形态学滤波 bw_img_in = bw_img.copy() kernel = cv.getStructuringElement(cv.MORPH_RECT, (5, 5)) morph_open_img = cv.morphologyEx(bw_img_in, cv.MORPH_OPEN, kernel, 3) # morph_close_img = cv.morphologyEx(bw_img, cv.MORPH_CLOSE, kernel, 3) # 白色比例分析 ratio_value = self.white_ratio_analyze(morph_open_img) # 连通区域分析，这里可设置阈值判断是否为缺陷区域 morph_open_img_label, stat = self.connect_component(morph_open_img) # print('分块序号：{0}，最大区域：{1}'.format(i, stat)) if stat[4] > self.thresh_value: result[i, 0] = 1 result[i, 1:] = stat return result, one_w if __name__ == '__main__': # 匹配文件名称 template_img_file = sys.argv[1].replace('\\','/') big_img_file = sys.argv[2].replace('\\','/') # print(template_img_file) # print(big_img_file) # print(int(sys.argv[3])) detector = DetectDefect() detector.set_big_image(big_img_file) detector.set_little_image(template_img_file) detector.set_roi_rect() # detector.set_thresh_value(thresh_value=6500) # 2000, 6500 detector.set_thresh_value(thresh_value=int(sys.argv[3])) # 2000, 6500 detector.show_roi_img() detector.split_patches() detector.defect_analyze_patchs() ``` ## 三、C#工程 ### 1. C#界面截图 ![image-20220326165532989](./image/1.png) ### 2. 控件变量 | 控件 | objectName | 属性设置 | 备注 | | :-----: | :--------: | :----------------------: | :--------------------------: | | Button | button1 | text: 运行 | 运行按钮 | | Button | button2 | text: 选择Template图片 | 打开文件夹，选择template图片 | | Button | button3 | text: 选择Sample图片 | 打开文件夹，选择sample图片 | | TextBox | textbox1 | enable：false | template图片文件路径显示 | | TextBox | textbox2 | enable：false | sample图片文件路径显示 | | TextBox | textbox3 | text： 6500 | thresh_value值 | | Label | label1 | text： thresh_value | | ### 3.功能代码 1. 打开文件夹，选择文件，获取文件路径 ```c# public string GetOpenFileDialogReturnFileFullName(bool multiSelect = false) { System.Windows.Forms.OpenFileDialog opdialog = new System.Windows.Forms.OpenFileDialog(); opdialog.Multiselect = multiSelect; opdialog.Filter = "bmp|*.bmp"; string files = null; if (opdialog.ShowDialog() == System.Windows.Forms.DialogResult.OK) { files = System.IO.Path.GetFullPath(opdialog.FileName); //绝对路径 } return files; } ``` 2. 打开cmd，运行命令行语句 ```c# public void RunCMDCommand(out string outPut, params string[] command) { using (Process pc = new Process()) { pc.StartInfo.FileName = "cmd.exe"; //pc.StartInfo.CreateNoWindow = false;//隐藏窗口运行 pc.StartInfo.RedirectStandardError = true;//重定向错误流 pc.StartInfo.RedirectStandardInput = true;//重定向输入流 pc.StartInfo.RedirectStandardOutput = true;//重定向输出流 pc.StartInfo.UseShellExecute = false; pc.Start(); int lenght = command.Length; foreach (string com in command) { pc.StandardInput.WriteLine(com);//输入CMD命令 } pc.StandardInput.WriteLine("exit");//结束执行，很重要的 pc.StandardInput.AutoFlush = true; outPut = pc.StandardOutput.ReadToEnd();//读取结果 pc.WaitForExit(); pc.Close(); } } ``` 3. 按钮点击 ```c# private void button1_Click(object sender, EventArgs e) { string resultStr = ""; thresh_value = textBox3.Text; string strInput1 = "conda activate myopencv"; string strInput2 = "python F:/PycharmProjects/MatchTemplate/demo_defect_detection.py " + template_dir + " " + sample_dir + " " + thresh_value; RunCMDCommand(out resultStr, strInput1, strInput2); Console.WriteLine(resultStr); } private void button2_Click(object sender, EventArgs e) { template_dir = GetOpenFileDialogReturnFileFullName(); textBox1.Text = template_dir; } private void button3_Click(object sender, EventArgs e) { sample_dir = GetOpenFileDialogReturnFileFullName(); textBox2.Text = sample_dir; } ```