def normalPressure(image, allinfo):
"""
:param image: ROI image
:param info: information for this meter
:return: value
"""
template = None
flag = None
info = None
if type(allinfo) is list:
for i in range(len(allinfo)): # todo 用传入的 多个模板 进行一个个的匹配，有一个匹配到则跳出循环
info = allinfo[i]
template, flag = meterFinderBySIFT(
image, info) # todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
else:
template, flag = meterFinderBySIFT(image, allinfo)
info = copy.deepcopy(allinfo)
if flag == 0:
print('not find template!!!')
result = scanPointer(template, info) # todo 调用 指针 识别函数
result = int(result * 1000) / 1000
return [result]
def colorIndicator(ROI, allinfo):
res = [0]
template = None
flag = None
info = None
if type(allinfo) is list:
for i in range(len(allinfo)):
info = allinfo[i]
template, flag = meterFinderBySIFT(
ROI, info) # todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
else:
template, flag = meterFinderBySIFT(ROI, allinfo)
info = copy.deepcopy(allinfo)
if flag == 0:
print('not find template!!!')
HSV = cv2.cvtColor(template, cv2.COLOR_BGR2HSV)
# color = [np.array([26, 43, 46]), np.array([34, 255, 255])]
# todo 根据颜色进行识别，结合实际图片
color = [np.array([11, 43, 46]), np.array([34, 255, 255])]
Lower = color[0]
Upper = color[1]
mask = cv2.inRange(HSV, Lower, Upper)
upmask = mask[int(0.25 * mask.shape[0]):int(0.5 * mask.shape[0]), :]
upmask = cv2.bitwise_and(np.ones(upmask.shape, np.uint8), upmask)
if np.sum(upmask) / upmask.shape[0] * upmask.shape[1] > 0.2:
res = [1]
return res
def onoffBattery(image, allinfo):
"""
识别电池屏内部的开关是向上还是向下
:param image:
:param info:
:return:
"""
template = None
flag = None
info = None
if type(allinfo) is list:
for i in range(len(allinfo)):
info = allinfo[i]
template, flag = meterFinderBySIFT(image, info) #todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
else:
template, flag = meterFinderBySIFT(image, allinfo)
info = copy.deepcopy(allinfo)
if flag == 0:
print('not find template!!!')
gray = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY)
start = ([info["startPoint"]["x"], info["startPoint"]["y"]])
end = ([info["endPoint"]["x"], info["endPoint"]["y"]])
center = ([info["centerPoint"]["x"], info["centerPoint"]["y"]])
# width = info["rectangle"]["width"]
# height = info["rectangle"]["height"]
width = 40
height = 80
fourth = (start[0] + end[0] - center[0], start[1] + end[1] - center[1])
pts1 = np.float32([start, center, end, fourth])
pts2 = np.float32([[0, 0], [width, 0], [width, height], [0, height]])
M = cv2.getPerspectiveTransform(pts1, pts2)
dst = cv2.warpPerspective(gray, M, (width, height))
# 自适应二值化
ret, thresh = cv2.threshold(dst, 0, 255, cv2.THRESH_OTSU)
# thresh = cv2.adaptiveThreshold(dst, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV, 17, 11)
#todo 比较上下区域的色度和，若上部区域比下部区域小，则说明开关是向上打开
upRegion = np.mean(thresh[:40])
downRegion = np.mean(thresh[40:])
if ifShow:
imgShow = np.hstack((dst, thresh))
cv2.imshow("ret", imgShow)
cv2.waitKey(0)
print(upRegion, downRegion)
return ["on"] if upRegion < downRegion else ["off"]
def switch(image, allinfo):
# delete_file.delete_files()
template = None
flag = None
info = None
if type(allinfo) is list:
for i in range(len(allinfo)):
info = allinfo[i]
template, flag = meterFinderBySIFT(
image, info) #todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
else:
template, flag = meterFinderBySIFT(image, allinfo)
info = copy.deepcopy(allinfo)
if flag == 0:
# print('not find template!!!')
pass
h, w, _ = template.shape
left = template[:, 0:w // 2].copy()
# todo 存储数据
#store_num = len(os.listdir("./store"))
#cv2.imwrite("./store/left_"+str(store_num)+".jpg",left)
left = cv2.resize(left, (40, 40))
left = np.expand_dims(left, 0) # todo 分为左右两个部分
right = template[:, w // 2:].copy()
# # todo 存储数据
# store_num = len(os.listdir("./store"))
# cv2.imwrite("./store/right_"+str(store_num)+".jpg",right)
right = cv2.resize(right, (40, 40))
right = np.expand_dims(right, 0)
net = switchNet() # todo 加载模型
net.load_state_dict(
torch.load("Algorithm/OCR/onoff/model/left_switch_net.pth"))
input = torch.tensor(left, dtype=torch.float32).permute((0, 3, 1, 2))
out = net(input)
out = out.detach().numpy() # 开 是 1 关 是 0
left_index = np.argmax(out)
net1 = switchNet() # todo 加载模型
net1.load_state_dict(
torch.load("Algorithm/OCR/onoff/model/right_switch_net.pth"))
input = torch.tensor(right, dtype=torch.float32).permute((0, 3, 1, 2))
out = net(input)
out = out.detach().numpy() # 开 是 1 关 是 0
right_index = np.argmax(out)
return [left_index, right_index]
def digitPressure(image, allinfo):
'''
:param image: 识别的图片
:param allinfo: 对应的 所有的 json文件
:return: 返回识别结果
'''
del_file(CUR_PATH) # todo 每次调用函数时，需要把上次存的图片删除，才能存储本次要识别的图片
# delete_file.delete_files()
template = None
flag = None
info = None
if type(allinfo) is list: # todo 用传入的 多个模板 进行一个个的匹配，有一个匹配到则跳出循环
for i in range(len(allinfo)):
info = allinfo[i]
# todo template 是根据模板 提取出图片上的区域
template, flag = meterFinderBySIFT(
image, info) #todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
else:
template, flag = meterFinderBySIFT(image, allinfo)
info = copy.deepcopy(allinfo)
if flag == 0:
# print('not find template!!!')
pass
# todo 这时 要识别的区域和对应的json文件已经准备好，开始进行数字识别
myRes = rgbRecognize(template, info)
merge = info["merge"] # todo 上下两行数字是否进行 合并
decimal = info["decimal"] # todo 小数点的位置
# todo 这个只是对识别结果进行一些格式上的处理
for i in range(len(myRes)):
temp = ""
for j, c in enumerate(myRes[i]):
if c != "?":
temp += c
elif j != 0:
temp += str(random.randint(0, 9))
myRes[i] = float(temp) if temp != "" else 0.0
if merge == 'true':
a = myRes[0] * 10**decimal[1] + myRes[1]
myRes.clear()
myRes.append(a)
return myRes
def springStatus(ROI, info):
# template = info['template']
# img = meterFinderByTemplate(ROI, template)
img = meterFinderBySIFT(ROI, info)
img = boxRectifier(img, info)
high, width = img.shape[:2]
i = 0
y = 2 * high // 5 # y坐标大约是2/5的高度
x = 0
while width - 6 - i > 0: # 比较相邻像素的三个通道的差值 从右往左比较
a1, a2 = img[y][width - 5 -
i][0].astype(int), img[y][width - 5 - i - 1][0].astype(
int) # 这里的5是为了防止扣取的模板不精确（防止边框抠出来的情况）
b1, b2 = img[y][width - 5 - i][1].astype(int), img[y][width - 5 - i -
1][1].astype(int)
c1, c2 = img[y][width - 5 - i][2].astype(int), img[y][width - 5 - i -
1][2].astype(int)
if abs(a1 - a2) < 20 and abs(b1 - b2) < 20 and abs(
c1 - c2) < 20: # 如果相邻像素值近似，继续循环
i += 1
else: # 不相似，输出当前点的x值
x = width - 5 - i - 1
break
# cv2.circle(img, (x, y), 2, (0, 0, 255), -1)
# cv2.imshow("ewe", img)
if x > 2 * width // 3:
return "on"
else:
return "off"
def onoffBatteryHardCode(image, info):
"""
识别电池屏内部的开关是向上还是向下
:param image: image
:param info: information
:return:
"""
meter = meterFinderBySIFT(image, info)
# print(meter.shape)
meter = cv2.cvtColor(meter, cv2.COLOR_BGR2GRAY)
meter = cv2.resize(meter, (200, 200))
# 截取中心区域
center = meter[60:120, 85:115]
# 自适应二值化
ret, thresh = cv2.threshold(center, 0, 255, cv2.THRESH_OTSU)
# 比较上下区域的色度和，若上部区域比下部区域小，则说明开关是向上打开
upRegion = np.sum(meter[:30])
downRegion = np.sum(meter[30:])
if ifShow:
imgShow = np.hstack((center, thresh))
cv2.imshow("ret", imgShow)
cv2.waitKey(0)
print(upRegion, downRegion)
return 1 if upRegion < downRegion else 0
def normalPressure(image, info):
"""
:param image: ROI image
:param info: information for this meter
:return: value
"""
meter = meterFinderBySIFT(image, info)
result = scanPointer(meter, info)
result = int(result * 1000) / 1000
return [result]
def oilTempreture(image, info):
"""
:param image: ROI image
:param info: information for this meter
:return: value
"""
start = np.array([info["startPoint"]["x"], info["startPoint"]["y"]])
end = np.array([info["endPoint"]["x"], info["endPoint"]["y"]])
center = np.array([info["centerPoint"]["x"], info["centerPoint"]["y"]])
meter = meterFinderBySIFT(image, info["template"])
h, w, _ = meter.shape
fixHeight = 300
fixWidth = int(meter.shape[1] / meter.shape[0] * fixHeight)
resizeCoffX = fixWidth / meter.shape[1]
meter = cv2.resize(meter, (fixWidth, fixHeight))
start = (start * resizeCoffX).astype(np.int16)
end = (end * resizeCoffX).astype(np.int16)
center = (center * resizeCoffX).astype(np.int16)
meter[int(0.6 * meter.shape[0]):] *= 0
mask_meter_red = color_detection(meter, red_range)
point_red = contours_check(mask_meter_red, center)
degree_red = AngleFactory.calPointerValueByOuterPoint(
start, end, center, point_red, info["startValue"], info["totalValue"])
cv2.destroyAllWindows()
mask_meter_white = color_detection(meter, white_range)
point_white = contours_check(mask_meter_white, center)
degree_white = AngleFactory.calPointerValueByOuterPoint(
start, end, center, point_white, info["startValue"],
info["totalValue"])
if ifShow:
print("white degree {:.2f}".format(degree_white))
print("red degree {:.2f}".format(degree_red))
cv2.circle(meter, (start[0], start[1]), 5, (0, 0, 255), -1)
cv2.circle(meter, (end[0], end[1]), 5, (0, 0, 255), -1)
cv2.circle(meter, (center[0], center[1]), 5, (0, 0, 255), -1)
cv2.circle(meter, (point_white[0], point_white[1]), 5, (255, 255, 255),
-1)
cv2.line(meter, (center[0], center[1]),
(point_white[0], point_white[1]), (255, 255, 255), 5)
cv2.line(meter, (center[0], center[1]), (point_red[0], point_red[1]),
(0, 0, 255), 5)
cv2.circle(meter, (point_red[0], point_red[1]), 5, (0, 0, 255), -1)
cv2.imshow("meter", meter)
cv2.waitKey(0)
return int(degree_red * 100) / 100, int(degree_white * 100) / 100
def colorIndicator(ROI, info):
res = 0
image = meterFinderBySIFT(ROI, info)
HSV = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
# color = [np.array([26, 43, 46]), np.array([34, 255, 255])]
color = [np.array([11, 43, 46]), np.array([34, 255, 255])]
Lower = color[0]
Upper = color[1]
mask = cv2.inRange(HSV, Lower, Upper)
upmask = mask[int(0.25 * mask.shape[0]):int(0.5 * mask.shape[0]), :]
upmask = cv2.bitwise_and(np.ones(upmask.shape, np.uint8), upmask)
if np.sum(upmask) / upmask.shape[0] * upmask.shape[1] > 0.2:
res = 1
return res
def checkFrame(net, image, info):
"""
判断图片的类型A，AB等
:param image: image
:param info: info
:return: 出现次序0.1.2.3
"""
start = ([info["startPoint"]["x"], info["startPoint"]["y"]])
end = ([info["endPoint"]["x"], info["endPoint"]["y"]])
center = ([info["centerPoint"]["x"], info["centerPoint"]["y"]])
width = info["rectangle"]["width"]
height = info["rectangle"]["height"]
widthSplit = info["widthSplit"]
heightSplit = info["heightSplit"]
characSplit = info["characSplit"]
# 计算数字表的矩形外框，并且拉直矫正
fourth = (start[0] + end[0] - center[0], start[1] + end[1] - center[1])
pts1 = np.float32([start, center, end, fourth])
pts2 = np.float32([[0, 0], [width, 0], [width, height], [0, height]])
M = cv2.getPerspectiveTransform(pts1, pts2)
template = meterFinderBySIFT(image, info)
template = cv2.cvtColor(template, cv2.COLOR_BGR2GRAY)
template = cv2.equalizeHist(template)
dst = cv2.warpPerspective(template, M, (width, height))
imgType = dst[characSplit[1][0]:characSplit[1][1],
characSplit[0][0]:characSplit[0][1]]
imgType = cv2.adaptiveThreshold(imgType, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY, 17, 11)
imgType = cv2.bitwise_not(imgType)
# orimage = imgType.copy()
imgType = torch.Tensor(np.array(imgType, dtype=np.float32))
imgType = torch.unsqueeze(imgType, 0)
imgType = torch.unsqueeze(imgType, 0)
type_probe = net.forward(imgType)
type_probe = type_probe.detach().numpy()
maxIndex = np.argmax(type_probe)
# debug
# cv2.imshow(str(chr(maxIndex+ord('A'))), orimage)
# type = str(chr(maxIndex+ord('A')))
# cv2.imwrite(os.path.join("video_result", type+info['name']+'_'+str(type_probe)+'.jpg'), orimage)
# cv2.waitKey(0)
return maxIndex
def onoffBattery(image, info):
"""
识别电池屏内部的开关是向上还是向下
:param image:
:param info:
:return:
"""
meter = meterFinderBySIFT(image, info)
gray = cv2.cvtColor(meter, cv2.COLOR_BGR2GRAY)
start = ([info["startPoint"]["x"], info["startPoint"]["y"]])
end = ([info["endPoint"]["x"], info["endPoint"]["y"]])
center = ([info["centerPoint"]["x"], info["centerPoint"]["y"]])
# width = info["rectangle"]["width"]
# height = info["rectangle"]["height"]
width = 40
height = 80
fourth = (start[0] + end[0] - center[0], start[1] + end[1] - center[1])
pts1 = np.float32([start, center, end, fourth])
pts2 = np.float32([[0, 0], [width, 0], [width, height], [0, height]])
M = cv2.getPerspectiveTransform(pts1, pts2)
dst = cv2.warpPerspective(gray, M, (width, height))
# 自适应二值化
ret, thresh = cv2.threshold(dst, 0, 255, cv2.THRESH_OTSU)
# thresh = cv2.adaptiveThreshold(dst, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY_INV, 17, 11)
# 比较上下区域的色度和，若上部区域比下部区域小，则说明开关是向上打开
upRegion = np.mean(thresh[:40])
downRegion = np.mean(thresh[40:])
if ifShow:
imgShow = np.hstack((dst, thresh))
cv2.imshow("ret", imgShow)
cv2.waitKey(0)
print(upRegion, downRegion)
return "on" if upRegion < downRegion else "off"
def digitPressure(image, info):
template = meterFinderBySIFT(image, info)
# 存储图片
if not os.path.exists("storeDigitData"):
os.mkdir("storeDigitData")
try:
os.mkdir("storeDigitData/thresh")
os.mkdir("storeDigitData/rgb")
except IOError:
pass
for i in range(11):
try:
os.mkdir("storeDigitData/thresh/" + str(i))
os.mkdir("storeDigitData/rgb/" + str(i))
except IOError:
continue
if 'rgb' in info and info['rgb']: # rgb as input
myRes = rgbRecognize(template, info)
else:
myRes = bitRecognize(template, info)
if info["digitType"] == "KWH":
myRes[0] = myRes[0][:4] + myRes.pop(1)
# 去除头部的非数字字符，同时将非头部的字符转为数字
for i in range(len(myRes)):
temp = ""
for j, c in enumerate(myRes[i]):
if c != "?":
temp += c
elif j != 0:
temp += str(random.randint(0, 9))
myRes[i] = float(temp) if temp != "" else 0.0
return myRes
def contactStatus(image, info):
"""
:param image:whole image
:param info:开关配置
:return: True代表O，False代表1
"""
template = meterFinderBySIFT(image, info)
template = cv2.GaussianBlur(template, (3, 3), 0)
# 获取检测区域
dst = boxRectifier(template, info)
debug = False
if debug:
cv2.imshow("image", image)
cv2.imshow("template", info['template'])
cv2.waitKey(0)
# 获取图像灰度矩阵
image = getMatrix(dst)
# 边缘检测
image = cv2.Canny(image, 10, 200)
return hasCircle(image)
def videoDigit(video, allinfo):
"""
:param video: VideoCapture Input
:param info: info
:return:
"""
net = characterNet(is_rgb=True) # todo 创建并加载 模型
net.load_state_dict(
torch.load("Algorithm/OCR/character/rbg2.pkl", map_location='cpu'))
pictures = getPictures(video) # todo 获得视频的帧，有少量重复帧
def emptyLsit():
return []
imagesDict = defaultdict(emptyLsit)
imagesDict2 = {
'A': -1,
'B': -1,
'C': -1
} # todo 防止没有识别出 ABC 三种帧 根据视频观察出 每隔120帧会变换利用这个特点进行第二种方法的编写
box = None
newinfo = None
newchange = None
saveframe = []
for i, frame in enumerate(pictures):
# res = digitPressure(frame, copy.deepcopy(info))
for j in range(len(allinfo)):
eachinfo = copy.deepcopy(allinfo[j])
template, flag = meterFinderBySIFT(
frame, eachinfo) # todo flag==0 则是没有匹配到，继续用下一个模板匹配
if flag == 0:
continue
else:
break
if flag == 0:
# print("not find template!!!")
pass
info = copy.deepcopy(eachinfo)
index, charimg = checkFrame(i, net, template,
eachinfo) # todo 识别当前的帧到底是属于哪一个类别
if index < 3:
res = rgbRecognize(template,
eachinfo) # todo 得出 数字的识别结果，存到对应的key值下
imagesDict[chr(index + ord('A'))] += [res]
if len(imagesDict['A']) != 0 and imagesDict2['B'] == -1:
imagesDict2['B'] = (i + 3) % len(pictures)
if len(imagesDict['B']) != 0 and imagesDict2['C'] == -1:
imagesDict2['C'] = (i + 3) % len(pictures)
if len(imagesDict['C']) != 0 and imagesDict2['A'] == -1:
imagesDict2['A'] = (i + 6) % len(pictures)
for key, val in imagesDict.items():
if len(val) == 0:
res = digitPressure(pictures[imagesDict2[key]], allinfo)
imagesDict[key] += [res]
imagesDict = getResult(imagesDict) #todo 对 结果做一些处理
return imagesDict
def readyStatus(img, info):
template = info['template']
# match_res =meterFinderByTemplate(img, info['template'])
match_res = meterFinderBySIFT(img, info)
image = boxRectifier(match_res, info)
# cv2.imshow('Image', image)
# cv2.waitKey(0)
# if image is dark enough, do gamma correction for enhancing dark details
if isDark(image):
max = np.max(image)
image = np.power(image / float(max), 1 / 3) * max
image = image.astype(np.uint8)
# cv2.imshow('Gamma', image)
# cv2.waitKey(0)
t_shape = template.shape[:2]
# image = meterFinderBySIFT(img, info)
orig = image.copy()
# minimum probability required to inspect a region
min_confidence = 0.5
# load the input image and grab the image dimensions
(H, W) = image.shape[:2]
# path to input EAST text detector
current_path = os.path.dirname(os.path.abspath(__file__)) + os.path.sep
model_path = os.path.abspath(current_path + "/frozen_east_text_detection.pb")
# set the new width and height and then determine the ratio in change
# for both the width and height
# image width should be multiple of 32, so do height
newW = (t_shape[0]) // 32 * 32
newH = (t_shape[1]) // 32 * 32
rW = W / float(newW)
rH = H / float(newH)
# resize the image and grab the new image dimensions
image = cv2.resize(image, (newW, newH))
(H, W) = image.shape[:2]
# define the two output layer names for the EAST detector model that
# we are interested -- the first is the output probabilities and the
# second can be used to derive the bounding box coordinates of text
layerNames = [
"feature_fusion/Conv_7/Sigmoid",
"feature_fusion/concat_3"]
# load the pre-trained EAST text detector
# print("[INFO] loading EAST text detector...")
net = cv2.dnn.readNet(model_path)
# construct a blob from the image and then perform a forward pass of
# the model to obtain the two output layer sets
blob = cv2.dnn.blobFromImage(image, 1.0, (W, H),
(123.68, 116.78, 103.94), swapRB=True, crop=False)
start = time.time()
net.setInput(blob)
(scores, geometry) = net.forward(layerNames)
end = time.time()
# show timing information on text prediction
# print("[INFO] text detection took {:.6f} seconds".format(end - start))
# grab the number of rows and columns from the scores volume, then
# initialize our set of bounding box rectangles and corresponding
# confidence scores
(numRows, numCols) = scores.shape[2:4]
rects = []
confidences = []
# loop over the number of rows
for y in range(0, numRows):
# extract the scores (probabilities), followed by the geometrical
# data used to derive potential bounding box coordinates that
# surround text
scoresData = scores[0, 0, y]
xData0 = geometry[0, 0, y]
xData1 = geometry[0, 1, y]
xData2 = geometry[0, 2, y]
xData3 = geometry[0, 3, y]
anglesData = geometry[0, 4, y]
# loop over the number of columns
for x in range(0, numCols):
# if our score does not have sufficient probability, ignore it
if scoresData[x] < min_confidence:
continue
# compute the offset factor as our resulting feature maps will
# be 4x smaller than the input image
(offsetX, offsetY) = (x * 4.0, y * 4.0)
# extract the rotation angle for the prediction and then
# compute the sin and cosine
angle = anglesData[x]
cos = np.cos(angle)
sin = np.sin(angle)
# use the geometry volume to derive the width and height of
# the bounding box
h = xData0[x] + xData2[x]
w = xData1[x] + xData3[x]
# compute both the starting and ending (x, y)-coordinates for
# the text prediction bounding box
endX = int(offsetX + (cos * xData1[x]) + (sin * xData2[x]))
endY = int(offsetY - (sin * xData1[x]) + (cos * xData2[x]))
startX = int(endX - w)
startY = int(endY - h)
# add the bounding box coordinates and probability score to
# our respective lists
rects.append((startX, startY, endX, endY))
confidences.append(scoresData[x])
if len(rects) > 0:
return "on"
else:
return "off"
def digitPressure(image, info):
template = meterFinderBySIFT(image, info)
template = cv2.GaussianBlur(template, (3, 3), 0)
# 读取标定信息
widthSplit = info["widthSplit"]
heightSplit = info["heightSplit"]
# 由标定点得到液晶区域
dst = boxRectifier(template, info)
# 灰度图
gray = cv2.cvtColor(dst, cv2.COLOR_BGR2GRAY)
# 针对不同的数字表类型进行不同的增强
if info["digitType"] != "TTC":
Blur = cv2.GaussianBlur(gray, (5, 5), 0)
Hist = cv2.equalizeHist(Blur)
thresh = cv2.adaptiveThreshold(Hist, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY_INV, 15, 11)
else:
thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C,
cv2.THRESH_BINARY, 55, 11)
# 存储图片
if not os.path.exists("storeDigitData"):
os.mkdir("storeDigitData")
if not os.path.exists("storeDigitData/digits"):
os.mkdir("storeDigitData/digits")
imgNum = int((len(os.listdir("storeDigitData/")) - 1) / 3)
cv2.imwrite("storeDigitData/" + str(imgNum) + "_dst.bmp", dst)
cv2.imwrite("storeDigitData/" + str(imgNum) + "_gray.bmp", gray)
cv2.imwrite("storeDigitData/" + str(imgNum) + "_thresh.bmp", thresh)
# 网络初始化
MyNet = newNet()
myRes = []
for i in range(len(heightSplit)):
split = widthSplit[i]
myNum = ""
for j in range(len(split) - 1):
if "decimal" in info.keys() and j == info["decimal"][i]:
myNum += "."
continue
# 得到分割的图片区域
img = thresh[heightSplit[i][0]:heightSplit[i][1],
split[j]:split[j + 1]]
# 增强
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 2))
img = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)
num = MyNet.recognizeNet(img)
myNum = myNum + num
# 存储图片
cv2.imwrite(
"storeDigitData/digits/{}_{}{}_p{}.bmp".format(
imgNum, i, j, num), img)
myRes.append(myNum)
if info["digitType"] == "KWH":
myRes[0] = myRes[0][:4] + myRes.pop(1)
# 去除头部的非数字字符，同时将非头部的字符转为数字
for i in range(len(myRes)):
temp = ""
for j, c in enumerate(myRes[i]):
if c != "?":
temp += c
elif j != 0:
temp += str(random.randint(0, 9))
myRes[i] = float(temp)
if ifShow:
cv2.imshow("rec", dst)
cv2.imshow("image", image)
print(myRes)
cv2.waitKey(0)
cv2.destroyAllWindows()
return myRes
