import time from PyQt5.QtGui import QPixmap, QImage from PyQt5.QtWidgets import QWidget, QFileDialog,QMainWindow, QLabel, QSizePolicy, QApplication, QAction, QHBoxLayout from PyQt5.QtCore import * from PyQt5 import QtCore, QtGui, QtWidgets import sys import traceback import ctypes as C import numpy as np import cv2 import os from test_0415_ui import Ui_MainWindow from pyueye import ueye import numpy as np import cv2 # 串口通信 import serial import time import threading import torch from Class.Camera import Camera_class from Class.Motor import Motor_class from Class.Pan import Pan_class from Class.Yolo import yolo_class class img_yolo(QtCore.QThread): sinOut = pyqtSignal(str) # 聲明一個帶字串參數的信號 def __init__(self, yolo_model, parent=None): super().__init__(parent) self.yolo = yolo_model def normalize_image(self, img): if img is not None: normalized_image = cv2.normalize(img, None, alpha=0, beta=255, norm_type=cv2.NORM_MINMAX) return normalized_image return None def process_image(self, img, filename): t1 = time.time() results = self.yolo.YoloDetect(img) t2 = time.time() print(f'推論時間 = {t2 - t1}') print(f'{filename}') # 如果有預測結果，則打印信心值 if results.pred[0] is not None and len(results.pred[0]) > 0: conf = results.pred[0][0, 4] # 信心值 print(f'信心值 = {conf}') if results.pred[0] is None or len(results.pred[0]) == 0: return predicted_classes = results.names[int(results.pred[0][0, -1])] if predicted_classes == 'fisheyes': processed_img = self.fisheyes_method(results,filename,img) elif predicted_classes == 'gel': # print('2') processed_img = self.gel_method(results,filename,img) elif predicted_classes == 'scratches': # print('3') processed_img = self.scratches_method(results,filename,img) # 魚眼處理 def fisheyes_method(self, results, filename, img): conf = results.pred[0][0, 4] predicted_classes = results.names[int(results.pred[0][0, -1])] box = results.pred[0][0, :4].cpu().numpy().astype(int) roi = img[box[1]:box[3], box[0]:box[2]] normalized_roi = self.normalize_image(roi) gray_roi = cv2.cvtColor(normalized_roi, cv2.COLOR_BGR2GRAY) clahe = cv2.createCLAHE(clipLimit=0.0, tileGridSize=(1, 1)) clahe_image = clahe.apply(gray_roi) _, binary_image = cv2.threshold(clahe_image, 200, 255, cv2.THRESH_BINARY) kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (6, 6)) closed_image = cv2.morphologyEx(binary_image, cv2.MORPH_CLOSE, kernel) contours, _ = cv2.findContours(closed_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) if len(contours) > 0: max_contour = max(contours, key=cv2.contourArea) if len(max_contour) > 5: ellipse = cv2.fitEllipse(max_contour) contour_image = roi.copy() cv2.ellipse(contour_image, ellipse, (0, 255, 0), 1) # center, axes, angle = ellipse # major_axis = max(axes) # longest_distance = 2 * major_axis * 5.5 / 1000 # 创建与图像相同大小的空白掩膜 mask = np.zeros_like(binary_image) # 在掩膜上绘制填充了的椭圆 cv2.ellipse(mask, ellipse, (255), thickness=cv2.FILLED) # 将掩膜应用于原始二值图像以获取椭圆内部的像素 ellipse_interior = cv2.bitwise_and(binary_image, binary_image, mask=mask) # 计算椭圆内部区域的像素总数，即面积 interior_area_pixels = np.sum(ellipse_interior == 255) interior_area_mm2 = interior_area_pixels * (5.5 ** 2) / 1000 # 将像素转换为平方毫米 # print(f"椭圆內部面積 (像素): {interior_area_pixels}") print(f"椭圆內部面積 (平方毫米): {interior_area_mm2}") # cv2.line(contour_image, (int(center[0] - major_axis / 2), int(center[1])), # (int(center[0] + major_axis / 2), int(center[1])), (0, 0, 255), 1) # print(f"最長距離：{longest_distance} mm") self.save_images(filename, roi, results, conf) # return contour_image # 凝膠處理 def gel_method(self, results, filename, img): conf = results.pred[0][0, 4] predicted_classes = results.names[int(results.pred[0][0, -1])] box = results.pred[0][0, :4].cpu().numpy().astype(int) roi = img[box[1]:box[3], box[0]:box[2]] normalized_roi = self.normalize_image(roi) gray_roi = cv2.cvtColor(normalized_roi, cv2.COLOR_BGR2GRAY) clahe = cv2.createCLAHE(clipLimit=0.0, tileGridSize=(1, 1)) clahe_image = clahe.apply(gray_roi) _, binary_image = cv2.threshold(clahe_image, 170, 255, cv2.THRESH_BINARY) # 使用 cv2.RETR_TREE 检索完整的轮廓层级 contours, _ = cv2.findContours(binary_image, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) total_area_pixels = 0 for i, contour in enumerate(contours): # 绘制填充了的轮廓 # cv2.drawContours(roi, [contour], -1, (255, 255, 255), thickness=cv2.FILLED) # 计算轮廓的面积并加到总面积中 contour_area_pixels = cv2.contourArea(contour) total_area_pixels += contour_area_pixels contour_area_mm2 = contour_area_pixels * (5.5 ** 2) / 1000 # 将像素转换为毫米 # print(f"轮廓 {i + 1} 的面积 (像素): {contour_area_pixels}") # print(f"轮廓 {i + 1} 的面积 (平方毫米): {contour_area_mm2}") # 计算总面积并将单位转换为毫米 total_area_mm2 = total_area_pixels * (5.5 ** 2) / 1000 # 将像素转换为毫米 # print(f"總面積 (像素): {total_area_pixels}") print(f"gel面積 (平方毫米): {total_area_mm2}") # 在原始图像上绘制轮廓 contour_image = roi.copy() cv2.drawContours(contour_image, contours, -1, (0, 255, 0), 1) self.save_images(filename, roi, results, conf) # if len(contours) > 0: # max_contour = max(contours, key=cv2.contourArea) # if len(max_contour) > 5: # ellipse = cv2.fitEllipse(max_contour) # contour_image = roi.copy() # cv2.ellipse(contour_image, ellipse, (0, 255, 0), 1) # center, axes, angle = ellipse # major_axis = max(axes) # longest_distance = 2 * major_axis * 5.5 / 1000 # cv2.line(contour_image, (int(center[0] - major_axis / 2), int(center[1])), # (int(center[0] + major_axis / 2), int(center[1])), (0, 0, 255), 1) # print(f"最長距離：{longest_distance} mm") # self.save_images(filename, contour_image, results, conf) # return contour_image # 刮痕處理 def scratches_method(self,results,filename,img): # conf = results.pred[0][0, 4] # predicted_classes = results.names[int(results.pred[0][0, -1])] # box = results.pred[0][0, :4].cpu().numpy().astype(int) # roi = img[box[1]:box[3], box[0]:box[2]] # normalized_roi = self.normalize_image(roi) # gray_roi = cv2.cvtColor(normalized_roi, cv2.COLOR_BGR2GRAY) # clahe = cv2.createCLAHE(clipLimit=0.0, tileGridSize=(1, 1)) # clahe_image = clahe.apply(gray_roi) # _, binary_image = cv2.threshold(clahe_image, 170, 255, cv2.THRESH_BINARY) # kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (6, 6)) # closed_image = cv2.morphologyEx(binary_image, cv2.MORPH_CLOSE, kernel) # contours, _ = cv2.findContours(closed_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # self.save_images(filename, normalized_roi, results, conf) conf = results.pred[0][0, 4] predicted_classes = results.names[int(results.pred[0][0, -1])] box = results.pred[0][0, :4].cpu().numpy().astype(int) roi = img[box[1]:box[3], box[0]:box[2]] # normalized_roi = self.normalize_image(roi) gray_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY) # clahe = cv2.createCLAHE(clipLimit=0.0, tileGridSize=(1, 1)) # clahe_image = clahe.apply(gray_roi) _, binary_image = cv2.threshold(gray_roi, 45, 240, cv2.THRESH_BINARY) kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (8, 8)) closed_image = cv2.morphologyEx(binary_image, cv2.MORPH_CLOSE, kernel) contours, _ = cv2.findContours(binary_image, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) total_area_pixels = 0 for i, contour in enumerate(contours): # 绘制填充了的轮廓 # cv2.drawContours(roi, [contour], -1, (255, 255, 255), thickness=cv2.FILLED) # 计算轮廓的面积并加到总面积中 contour_area_pixels = cv2.contourArea(contour) total_area_pixels += contour_area_pixels contour_area_mm2 = contour_area_pixels * (5.5 ** 2) / 1000 # 将像素转换为毫米 # print(f"轮廓 {i + 1} 的面积 (像素): {contour_area_pixels}") # print(f"轮廓 {i + 1} 的面积 (平方毫米): {contour_area_mm2}") # 计算总面积并将单位转换为毫米 total_area_mm2 = total_area_pixels * (5.5 ** 2) / 1000 # 将像素转换为毫米 # print(f"總面積 (像素): {total_area_pixels}") print(f"scratches面積 (平方毫米): {total_area_mm2}") # 在原始图像上绘制轮廓 contour_image = roi.copy() cv2.drawContours(contour_image, contours, -1, (0, 255, 0), 1) self.save_images(filename, roi, results, conf) # if len(contours) > 0: # max_contour = max(contours, key=cv2.contourArea) # if len(max_contour) > 5: # ellipse = cv2.fitEllipse(max_contour) # contour_image = roi.copy() # cv2.ellipse(contour_image, ellipse, (0, 255, 0), 1) # center, axes, angle = ellipse # major_axis = max(axes) # longest_distance = 2 * major_axis * 5.5 / 1000 # cv2.line(contour_image, (int(center[0] - major_axis / 2), int(center[1])), # (int(center[0] + major_axis / 2), int(center[1])), (0, 0, 255), 1) # print(f"最長距離：{longest_distance} mm") # self.save_images(filename, contour_image, results, conf) # return contour_image def save_images(self, filename, contour_image, results, conf): if conf > 0.8: base_name, ext = os.path.splitext(filename) new_file_name = f'{base_name}_2{ext}' print(f'Saving {new_file_name}') cv2.imwrite(f'img_2/{new_file_name}', contour_image) # if conf > 0.6: # resized_frame_high_conf = cv2.resize(results.render()[0], (1280, 1280)) # print(f'Saving {filename}') # cv2.imwrite(f'img_2/{filename}', resized_frame_high_conf) def run(self): print('run') self.img_list = [] self.filename_list = [] for filename in os.listdir(r"./image_folder"): if filename.endswith('.jpg') or filename.endswith('.bmp'): img = cv2.imread('image_folder' + "/" + filename) self.img_list.append(img) self.filename_list.append(filename) print(f'img讀取完成') for i in range(0, len(self.filename_list)): img = self.img_list[i] file_name = self.filename_list[i] self.process_image(img, file_name) def img_to_view(img):#原圖 img= cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # QT顏色顯示轉換 Ny, Nx, _ = img.shape img = QtGui.QImage(img.data, Nx, Ny,Nx*3, QtGui.QImage.Format_RGB888) #須改格式 return img class MainWindow(QMainWindow,Ui_MainWindow): def __init__(self, parent=None): #按鍵設定 super(MainWindow, self).__init__(parent) self.setupUi(self) # self.view_1.setScaledContents(True) # self.view_2.setScaledContents(True) # self.view_3.setScaledContents(True) # self.view_4.setScaledContents(True) self.bt_1.clicked.connect(self.bt_1_clicked) # self.bt_2.clicked.connect(self.bt_2_clicked) # 連接按鈕到新的功能 self.bt_3.clicked.connect(self.bt_3_clicked) # self.cam = Camera_class() # self.cam.start() # self.cam.rawdata.connect(self.show_img) # # self.run_flag = True # self.com5 = Motor_class() # self.com5.start() # self.com5.sinOut.connect(self.com5_str) # # # self.com4 = Pan_class() # self.com4.start() # self.com4.sinOut.connect(self.com4_str) # self.count = 0 # self.img_count=0 self.yolo = yolo_class('weight/best.pt') #self.yolo.start() # self.img_test_2() def show_img(self,img): self.img = img h,w,_ = self.img.shape img= img_to_view(img) self.view_1.setPixmap(QtGui.QPixmap.fromImage(img)) def bt_1_clicked(self): self.time_2= time.time() self.com5.start_run() self.run_flag = True self.time_1 = time.time() def com5_str(self,str_1): if str_1 =='move': self.count = self.count+1 if self.count< 6: print(self.count) if str_1 == 'move': self.img_count = 0 self.com4.move() if self.count == 5: self.run_flag = False if str_1 =='2': # self.img_test() #self.time_1 = time.time() #print(f's={self.time_1-self.time_2}') # print('save_img') self.img_count = self.img_count+1 cv2.imwrite(f'image_folder\\{str(self.count)}_{self.img_count}.jpg',self.img) #self.time_2= time.time() def com4_str(self,str_1): if str_1 =='move_end' and self.run_flag == True: self.com5.start_run() if str_1 == 'Motor emergency stop 010': print('010') self.run_flag = False self.count = 0 time_2 = time.time() print(f'all_time - {time_2 - self.time_1}') self.bt_3_clicked() def img_test_2(self): img_list=[] for filename in os.listdir(r"./image_folder"): if filename.endswith('.jpg'): img = cv2.imread('image_folder' + "/" + filename) img_list.append(img) # self.img_test(img, filename) print(f'img讀取完成') run_list=[] for img in img_list: run_list.append(img_yolo( self.yolo,img)) for i in range (0,len(run_list)): run_list[i].start() def bt_3_clicked(self): self.T1 = img_yolo(self.yolo) self.T1.finished.connect(self.on_yolo_finished) # Connect to finished signal self.T1.start() def on_yolo_finished(self): img_confidence = {} # 用于存储图像的置信度值 # 从img_2文件夹加载图像 for filename in os.listdir(r"./img_2"): if filename.endswith('.jpg'): img_path = os.path.join("./img_2", filename) img = cv2.imread(img_path) img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) h, w, _ = img_rgb.shape qimg = QtGui.QImage(img_rgb.data, w, h, w * 3, QtGui.QImage.Format_RGB888) # 执行YOLO检测并获取置信度值 results = self.yolo.YoloDetect(img) if results.pred[0] is not None and len(results.pred[0]) > 0: conf = results.pred[0][0, 4] # 置信度值 else: conf = 0 img_confidence[filename] = conf # 按置信度值降序排列图像 sorted_imgs = sorted(img_confidence.items(), key=lambda x: x[1], reverse=True) count = 0 # 计算显示的图像数量 # 基于置信度值在view_2、view_3和view_4中显示图像 for filename, _ in sorted_imgs: img_path = os.path.join("./img_2", filename) img = cv2.imread(img_path) img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) h, w, _ = img_rgb.shape qimg = QtGui.QImage(img_rgb.data, w, h, w * 3, QtGui.QImage.Format_RGB888) if not self.view_2.pixmap(): self.view_2.setPixmap(QtGui.QPixmap.fromImage(qimg)) count += 1 elif not self.view_3.pixmap(): self.view_3.setPixmap(QtGui.QPixmap.fromImage(qimg)) count += 1 elif not self.view_4.pixmap(): self.view_4.setPixmap(QtGui.QPixmap.fromImage(qimg)) count += 1 # 如果显示了3张图像，则退出循环 if count == 3: break if __name__ == "__main__": app = QApplication(sys.argv) window = MainWindow() window.show() sys.exit(app.exec_())