遥感数据集

1. UC Merced Land-Use Data Set图像像素大小为256*256，总包含21类场景图像，每一类有100张，共2100张。http://weegee.vision.ucmerced.edu/datasets/landuse.html2. WHU-RS19 Data Set图像像素大小为600*600，总包含19类场景图像，每一类大概50张，共1005张。/download/u010656161/101534103. SIRI-WHU Data Set图像像素大小为200*200，总包含12类场景图像，每一类有200张，共2400张。4. RSSCN7 Data Set图像像素大小为400*400，总包含7类场景图像，每一类有400张，共2800张。5. RSC11 Data Set图像像素大小为512*512，总包含11类场景图像，每一类大概100张，共1232张。6. NWPU-RESISC45 Data Set/people/JunweiHan/NWPU-RESISC45.html7. Road andBuilding Detection Data Sethttps://www.cs.toronto.edu/~vmnih/data/8. DOTA: A Large-scale Dataset for Object Detection inAerial Imageshttp://captain./DOTAweb/index.html9. DeepGlobe卫星图像地表解析(道路提取、建筑物检测、地标分类)挑战赛/challenge.html

CVPR 挑战赛

网址：/leaderboard.html

基于深度学习的影像地图道路提取

网络结构：D-LinkNet - LinkNet with Pretrained Encoder and Dilated Convolution for HighResolution Satellite Imagery Road Extraction

与Linknet的区别：增加了下图中的B部分，即扩张卷积层，通过多个卷积层信息的叠加，可以最大化地增大感受视野范围，同时利用ResNet34来替换掉18。

预测：由于实验需要的图片需要长与宽一致，因此在谷歌地图上找了256*256（论文声称支持1024*1024）的瓦片来进行实验，发现必须找到16级以上的瓦片才可以实现检测，可能训练数据集中需要道路的宽度具有一定的长度才可以。

基于SegNet和U-Net的遥感图像语义分割

Blog：/real_myth/article/details/79432456

GitHub：/AstarLight/Satellite-Segmentation（Satellite_Image_Segmentation_BY_SegNet_UNet）

步骤：

Segmented by SegNetSegmented by U-NetModel Emsamble：SegNet + U-Net

数据集

数据下载：/s/1i6oMukH（密码：yqj2）|/s/1FwHkvp2esvhyOx1eSZfkog（密码：fqnw)

|-test 测试图片|-train SegNet训练集|----label 标记图|----src 遥感图|-unet_buildings UNet训练集|----label 标记图|----src 遥感图

数据集来自CCF大数据比赛提供的数据(中国南方某城市的高清遥感图像)，是小数据集，里面包含了5张带标注的大尺寸RGB遥感图像(尺寸范围从3000×3000到6000×6000)

里面一共标注了4类物体：植被(标记1)、建筑(标记2)、水体(标记3)、道路(标记4)以及其他(标记0)。其中，耕地、林地、草地均归为植被类。更多数据介绍可以参看这里(/#/competitions/270/data-intro)

训练图片及其标记图片可视化后的效果如下：蓝色-水体，黄色-房屋，绿色-植被，棕色-马路

问题：Label可视化-原始数据集里的训练集图片采用十六位的，图片浏览器显示全黑（一般图片浏览器无法显示16位图）

解决： 将深度16位的图片转为8位（比如，Matlab下：im2 = uint8(im1))

数据处理

原始图像：5张大尺寸的遥感图像(尺寸各不相同)

随机切割：随机生成x,y坐标，然后抠出该坐标下256*256的小图

# 执行切割 - UNet训练集python ./unet/gen_dataset.py

数据增强(Keras自带的数据增广函数/)：

原图和label图都需要旋转：90度、180度、270度原图和label图都需要做沿y轴的镜像操作原图做模糊操作原图做光照调整操作原图做增加噪声操作(高斯噪声、椒盐噪声)

# OpenCV编写的相应的增强函数

img_w = 256 img_h = 256 image_sets = ['1.png','2.png','3.png','4.png','5.png']def gamma_transform(img, gamma):gamma_table = [np.power(x / 255.0, gamma) * 255.0 for x in range(256)]gamma_table = np.round(np.array(gamma_table)).astype(np.uint8)return cv2.LUT(img, gamma_table)def random_gamma_transform(img, gamma_vari):log_gamma_vari = np.log(gamma_vari)alpha = np.random.uniform(-log_gamma_vari, log_gamma_vari)gamma = np.exp(alpha)return gamma_transform(img, gamma)def rotate(xb,yb,angle):M_rotate = cv2.getRotationMatrix2D((img_w/2, img_h/2), angle, 1)xb = cv2.warpAffine(xb, M_rotate, (img_w, img_h))yb = cv2.warpAffine(yb, M_rotate, (img_w, img_h))return xb,ybdef blur(img):img = cv2.blur(img, (3, 3));return imgdef add_noise(img):for i in range(200): #添加点噪声temp_x = np.random.randint(0,img.shape[0])temp_y = np.random.randint(0,img.shape[1])img[temp_x][temp_y] = 255return imgdef data_augment(xb,yb):if np.random.random() < 0.25:xb,yb = rotate(xb,yb,90)if np.random.random() < 0.25:xb,yb = rotate(xb,yb,180)if np.random.random() < 0.25:xb,yb = rotate(xb,yb,270)if np.random.random() < 0.25:xb = cv2.flip(xb, 1) # flipcode>0：沿y轴翻转yb = cv2.flip(yb, 1)if np.random.random() < 0.25:xb = random_gamma_transform(xb,1.0)if np.random.random() < 0.25:xb = blur(xb)if np.random.random() < 0.2:xb = add_noise(xb)return xb,ybdef creat_dataset(image_num = 100000, mode = 'original'):print('creating dataset...')image_each = image_num / len(image_sets)g_count = 0for i in tqdm(range(len(image_sets))):count = 0src_img = cv2.imread('./data/src/' + image_sets[i]) # 3 channelslabel_img = cv2.imread('./data/label/' + image_sets[i],cv2.IMREAD_GRAYSCALE) # single channelX_height,X_width,_ = src_img.shapewhile count < image_each:random_width = random.randint(0, X_width - img_w - 1)random_height = random.randint(0, X_height - img_h - 1)src_roi = src_img[random_height: random_height + img_h, random_width: random_width + img_w,:]label_roi = label_img[random_height: random_height + img_h, random_width: random_width + img_w]if mode == 'augment':src_roi,label_roi = data_augment(src_roi,label_roi)visualize = np.zeros((256,256)).astype(np.uint8)visualize = label_roi *50cv2.imwrite(('./aug/train/visualize/%d.png' % g_count),visualize)cv2.imwrite(('./aug/train/src/%d.png' % g_count),src_roi)cv2.imwrite(('./aug/train/label/%d.png' % g_count),label_roi)count += 1 g_count += 1

经过以上数据增强操作后，可得到了较大的训练集：100000张256*256的图片

卷积神经网络模型训练

图像语义分割任务-模型选择：FCN、U-Net、SegNet、DeepLab、RefineNet、Mask Rcnn、Hed Net

SegNet -网络结构清晰易懂，训练快

# 执行训练 - 修改filepath为segnet训练集路径python segnet_train.py --model segnet.h5 # --model后指定保存的模型名

网络结构定义：编码器-解码器（做语义分割时通常在末端加入CRF模块做后处理，旨在进一步精修边缘的分割结果）

def SegNet(): model = Sequential() #encoder model.add(Conv2D(64,(3,3),strides=(1,1),input_shape=(3,img_w,img_h),padding='same',activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(64,(3,3),strides=(1,1),padding='same',activation='relu')) model.add(BatchNormalization()) model.add(MaxPooling2D(pool_size=(2,2))) #(128,128) model.add(Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(MaxPooling2D(pool_size=(2, 2))) #(64,64) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(MaxPooling2D(pool_size=(2, 2))) #(32,32) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(MaxPooling2D(pool_size=(2, 2))) #(16,16) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(MaxPooling2D(pool_size=(2, 2))) #(8,8) #decoder model.add(UpSampling2D(size=(2,2))) #(16,16) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(UpSampling2D(size=(2, 2))) #(32,32) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(512, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(UpSampling2D(size=(2, 2))) #(64,64) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(256, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(UpSampling2D(size=(2, 2))) #(128,128) model.add(Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(128, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(UpSampling2D(size=(2, 2))) #(256,256) model.add(Conv2D(64, (3, 3), strides=(1, 1), input_shape=(3,img_w, img_h), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(64, (3, 3), strides=(1, 1), padding='same', activation='relu')) model.add(BatchNormalization()) model.add(Conv2D(n_label, (1, 1), strides=(1, 1), padding='same')) model.add(Reshape((n_label,img_w*img_h))) #axis=1和axis=2互换位置，等同于np.swapaxes(layer,1,2) model.add(Permute((2,1))) model.add(Activation('softmax')) pile(loss='categorical_crossentropy',optimizer='sgd',metrics=['accuracy']) model.summary() return model

划分数据集：读入数据集，选择的验证集大小是训练集的0.25

def get_train_val(val_rate = 0.25):train_url = [] train_set = []val_set = []for pic in os.listdir(filepath + 'src'):train_url.append(pic)random.shuffle(train_url)total_num = len(train_url)val_num = int(val_rate * total_num)for i in range(len(train_url)):if i < val_num:val_set.append(train_url[i]) else:train_set.append(train_url[i])return train_set,val_set# data for training def generateData(batch_size,data=[]): #print 'generateData...'while True: train_data = [] train_label = [] batch = 0 for i in (range(len(data))): url = data[i]batch += 1 #print (filepath + 'src/' + url)#img = load_img(filepath + 'src/' + url, target_size=(img_w, img_h)) img = load_img(filepath + 'src/' + url)img = img_to_array(img) # print img# print img.shape train_data.append(img) #label = load_img(filepath + 'label/' + url, target_size=(img_w, img_h),grayscale=True)label = load_img(filepath + 'label/' + url, grayscale=True)label = img_to_array(label).reshape((img_w * img_h,)) # print label.shape train_label.append(label) if batch % batch_size==0: #print 'get enough bacth!\n'train_data = np.array(train_data) train_label = np.array(train_label).flatten() train_label = labelencoder.transform(train_label) train_label = to_categorical(train_label, num_classes=n_label) train_label = train_label.reshape((batch_size,img_w * img_h,n_label)) yield (train_data,train_label) train_data = [] train_label = [] batch = 0 # data for validation def generateValidData(batch_size,data=[]): #print 'generateValidData...'while True: valid_data = [] valid_label = [] batch = 0 for i in (range(len(data))): url = data[i]batch += 1 #img = load_img(filepath + 'src/' + url, target_size=(img_w, img_h))img = load_img(filepath + 'src/' + url)#print img#print (filepath + 'src/' + url)img = img_to_array(img) # print img.shape valid_data.append(img) #label = load_img(filepath + 'label/' + url, target_size=(img_w, img_h),grayscale=True)label = load_img(filepath + 'label/' + url, grayscale=True)label = img_to_array(label).reshape((img_w * img_h,)) # print label.shape valid_label.append(label) if batch % batch_size==0: valid_data = np.array(valid_data) valid_label = np.array(valid_label).flatten() valid_label = labelencoder.transform(valid_label) valid_label = to_categorical(valid_label, num_classes=n_label) valid_label = valid_label.reshape((batch_size,img_w * img_h,n_label)) yield (valid_data,valid_label) valid_data = [] valid_label = [] batch = 0

训练：batch size定为16，epoch定为30，每次都存储最佳model(save_best_only=True)，并且在训练结束时绘制loss/acc曲线，并存储起来

def train(args): EPOCHS = 30BS = 16model = SegNet() modelcheck = ModelCheckpoint(args['model'],monitor='val_acc',save_best_only=True,mode='max') callable = [modelcheck] train_set,val_set = get_train_val()train_numb = len(train_set) valid_numb = len(val_set) print ("the number of train data is",train_numb) print ("the number of val data is",valid_numb)H = model.fit_generator(generator=generateData(BS,train_set),steps_per_epoch=train_numb//BS,epochs=EPOCHS,verbose=1, validation_data=generateValidData(BS,val_set),validation_steps=valid_numb//BS,callbacks=callable,max_q_size=1) # plot the training loss and accuracyplt.style.use("ggplot")plt.figure()N = EPOCHSplt.plot(np.arange(0, N), H.history["loss"], label="train_loss")plt.plot(np.arange(0, N), H.history["val_loss"], label="val_loss")plt.plot(np.arange(0, N), H.history["acc"], label="train_acc")plt.plot(np.arange(0, N), H.history["val_acc"], label="val_acc")plt.title("Training Loss and Accuracy on SegNet Satellite Seg")plt.xlabel("Epoch #")plt.ylabel("Loss/Accuracy")plt.legend(loc="lower left")plt.savefig(args["plot"])

预测- 预测整张遥感图像：1）训练模型时选择的图片输入是256×256，所以预测时也要采用256×256的图片尺寸送进模型预测；2）将预测好的小图重新拼接成一个大图 -先给大图做padding 0操作，得到一副padding过的大图，同时我们也生成一个与该图一样大的全0图A，把图像的尺寸补齐为256的倍数，然后以256为步长切割大图，依次将小图送进模型预测，预测好的小图则放在A的相应位置上，依次进行，最终得到预测好的整张大图（即A），再做图像切割，切割成原先图片的尺寸，完成整个预测流程

# 执行预测 - 修改待预测的图片的路径python segnet_predict.py

def predict(args):# load the trained convolutional neural networkprint("[INFO] loading network...")model = load_model(args["model"])stride = args['stride']for n in range(len(TEST_SET)):path = TEST_SET[n]#load the imageimage = cv2.imread('./test/' + path)# pre-process the image for classification#image = image.astype("float") / 255.0#image = img_to_array(image)h,w,_ = image.shapepadding_h = (h//stride + 1) * stride padding_w = (w//stride + 1) * stridepadding_img = np.zeros((padding_h,padding_w,3),dtype=np.uint8)padding_img[0:h,0:w,:] = image[:,:,:]padding_img = padding_img.astype("float") / 255.0padding_img = img_to_array(padding_img)print 'src:',padding_img.shapemask_whole = np.zeros((padding_h,padding_w),dtype=np.uint8)for i in range(padding_h//stride):for j in range(padding_w//stride):crop = padding_img[:3,i*stride:i*stride+image_size,j*stride:j*stride+image_size]_,ch,cw = crop.shapeif ch != 256 or cw != 256:print 'invalid size!'continuecrop = np.expand_dims(crop, axis=0)#print 'crop:',crop.shapepred = model.predict_classes(crop,verbose=2) pred = labelencoder.inverse_transform(pred[0]) #print (np.unique(pred)) pred = pred.reshape((256,256)).astype(np.uint8)#print 'pred:',pred.shapemask_whole[i*stride:i*stride+image_size,j*stride:j*stride+image_size] = pred[:,:]cv2.imwrite('./predict/pre'+str(n+1)+'.png',mask_whole[0:h,0:w])

预测效果图：

问题：预测图Mask-Label可视化 -每类物体对应的标签的值都是1到5，都接近黑色

解决：/AstarLight/Satellite-Segmentation/blob/master/draw_lables.cpp

import cv2import numpy as npALL = 0VEGETATION = 1ROAD = 4BUILDING = 2WATER = 3TEST_SET = ['1.png','2.png','3.png']Mask_Set = ['pre1.png','pre2.png','pre3.png']for n in range(len(TEST_SET)):print(n)path = TEST_SET[n]mask_path = Mask_Set[n]src = cv2.imread('../data/remote_sensing_image/test/' + path)mask = cv2.imread('./predict/'+mask_path)print(np.shape(mask))h,w,_ = src.shapefor i in range(0, h):for j in range(0, w):if (mask[i, j, 0] == VEGETATION):src[i, j, 0] = 159src[i, j, 1] = 255src[i, j, 2] = 84if (mask[i, j, 0] == ROAD):src[i, j, 0] = 38src[i, j, 1] = 71src[i, j, 2] = 139if (mask[i, j, 0] == BUILDING):src[i, j, 0] = 34src[i, j, 1] = 180src[i, j, 2] = 238if (mask[i, j, 0] == WATER):src[i, j, 0] = 255src[i, j, 1] = 191src[i, j, 2] = 0cv2.imwrite('./predict/stack' + str(n + 1) + '.png', src)

问题：拼接痕迹过于明显

解决：缩小切割时的滑动步伐，比如把切割步伐改为128，那么拼接时就会有一半的图像发生重叠，这样做可以尽可能地减少拼接痕迹

U-Net - 小数据集也能训练出好的模型，训练快

# 执行训练python unet_train.py --model unet_buildings20.h5 --data ./unet_train/buildings/ # --data后指定UNet训练集路径

网络结构定义：整个呈现U形，故起名U-Net

1）四类物体 - 多分类模型 -直接4分类

2）每一类物体 -二分类模型 - 得到4张预测图，再做预测图叠加，合并成一张完整的包含4类的预测图（loss function = binary_crossentropy 训练二分类模型）

def unet():inputs = Input((3, img_w, img_h))conv1 = Conv2D(32, (3, 3), activation="relu", padding="same")(inputs)conv1 = Conv2D(32, (3, 3), activation="relu", padding="same")(conv1)pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)conv2 = Conv2D(64, (3, 3), activation="relu", padding="same")(pool1)conv2 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv2)pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)conv3 = Conv2D(128, (3, 3), activation="relu", padding="same")(pool2)conv3 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv3)pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)conv4 = Conv2D(256, (3, 3), activation="relu", padding="same")(pool3)conv4 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv4)pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)conv5 = Conv2D(512, (3, 3), activation="relu", padding="same")(pool4)conv5 = Conv2D(512, (3, 3), activation="relu", padding="same")(conv5)up6 = concatenate([UpSampling2D(size=(2, 2))(conv5), conv4], axis=1)conv6 = Conv2D(256, (3, 3), activation="relu", padding="same")(up6)conv6 = Conv2D(256, (3, 3), activation="relu", padding="same")(conv6)up7 = concatenate([UpSampling2D(size=(2, 2))(conv6), conv3], axis=1)conv7 = Conv2D(128, (3, 3), activation="relu", padding="same")(up7)conv7 = Conv2D(128, (3, 3), activation="relu", padding="same")(conv7)up8 = concatenate([UpSampling2D(size=(2, 2))(conv7), conv2], axis=1)conv8 = Conv2D(64, (3, 3), activation="relu", padding="same")(up8)conv8 = Conv2D(64, (3, 3), activation="relu", padding="same")(conv8)up9 = concatenate([UpSampling2D(size=(2, 2))(conv8), conv1], axis=1)conv9 = Conv2D(32, (3, 3), activation="relu", padding="same")(up9)conv9 = Conv2D(32, (3, 3), activation="relu", padding="same")(conv9)conv10 = Conv2D(n_label, (1, 1), activation="sigmoid")(conv9)#conv10 = Conv2D(n_label, (1, 1), activation="softmax")(conv9) model = Model(inputs=inputs, outputs=conv10)pile(optimizer='Adam', loss='binary_crossentropy', metrics=['accuracy'])return model

划分数据集：读取数据

# data for training def generateData(batch_size,data=[]): #print 'generateData...'while True: train_data = [] train_label = [] batch = 0 for i in (range(len(data))): url = data[i]batch += 1 img = load_img(filepath + 'src/' + url)img = img_to_array(img) train_data.append(img) label = load_img(filepath + 'label/' + url, grayscale=True) label = img_to_array(label)#print label.shape train_label.append(label) if batch % batch_size==0: #print 'get enough bacth!\n'train_data = np.array(train_data) train_label = np.array(train_label) yield (train_data,train_label) train_data = [] train_label = [] batch = 0 # data for validation def generateValidData(batch_size,data=[]): #print 'generateValidData...'while True: valid_data = [] valid_label = [] batch = 0 for i in (range(len(data))): url = data[i]batch += 1 img = load_img(filepath + 'src/' + url)#print imgimg = img_to_array(img) # print img.shape valid_data.append(img) label = load_img(filepath + 'label/' + url, grayscale=True)valid_label.append(label) if batch % batch_size==0: valid_data = np.array(valid_data) valid_label = np.array(valid_label) yield (valid_data,valid_label) valid_data = [] valid_label = [] batch = 0

训练：指定输出model名字和训练集位置

python unet.py --model unet_buildings20.h5 --data ./unet_train/buildings/

预测-预测单张遥感图像：分别使用4个模型做预测，得到4张mask（比如下图是用训练好的buildings模型预测的结果），再将4张mask合并成1张 - 通过观察每一类的预测结果，根据不同类物体的预测准确率，给4类mask图排优先级（building>water>road>vegetation），当遇到一个像素点，4个mask图都说是属于自己类别的标签时，就可以根据先前定义好的优先级，把该像素的标签定为优先级最高的标签

# 执行预测python unet_predict.py

def combind_all_mask():for mask_num in tqdm(range(3)):if mask_num == 0:final_mask = np.zeros((5142,5664),np.uint8)#生成一个全黑全0图像,图片尺寸与原图相同elif mask_num == 1:final_mask = np.zeros((2470,4011),np.uint8)elif mask_num == 2:final_mask = np.zeros((6116,3356),np.uint8)#final_mask = cv2.imread('final_1_8bits_predict.png',0)if mask_num == 0:mask_pool = mask1_poolelif mask_num == 1:mask_pool = mask2_poolelif mask_num == 2:mask_pool = mask3_poolfinal_name = img_sets[mask_num]for idx,name in enumerate(mask_pool):img = cv2.imread('./predict_mask/'+name,0)height,width = img.shapelabel_value = idx+1 #coressponding labels valuefor i in tqdm(range(height)): #priority:building>water>road>vegetationfor j in range(width):if img[i,j] == 255:if label_value == 2:final_mask[i,j] = label_valueelif label_value == 3 and final_mask[i,j] != 2:final_mask[i,j] = label_valueelif label_value == 4 and final_mask[i,j] != 2 and final_mask[i,j] != 3:final_mask[i,j] = label_valueelif label_value == 1 and final_mask[i,j] == 0:final_mask[i,j] = label_valuecv2.imwrite('./final_result/'+final_name,final_mask) print 'combinding mask...'combind_all_mask()

模型融合

集成学习：两个模型 + 模型采取不同参数训练 -得到很多预测MASK图 -对每张结果图的每个像素点采取投票表决

少数服从多数的投票表决：对每张图相应位置的像素点的类别进行预测，票数最多的类别即为该像素点的类别 -可以很好地去掉一些明显分类错误的像素点，很大程度上改善模型的预测能力

import numpy as npimport cv2import argparseRESULT_PREFIXX = ['./result1/','./result2/','./result3/']# each mask has 5 classes: 0~4def vote_per_image(image_id):result_list = []for j in range(len(RESULT_PREFIXX)):im = cv2.imread(RESULT_PREFIXX[j]+str(image_id)+'.png',0)result_list.append(im)# each pixelheight,width = result_list[0].shapevote_mask = np.zeros((height,width))for h in range(height):for w in range(width):record = np.zeros((1,5))for n in range(len(result_list)):mask = result_list[n]pixel = mask[h,w]#print('pix:',pixel)record[0,pixel]+=1label = record.argmax()#print(label)vote_mask[h,w] = labelcv2.imwrite('vote_mask'+str(image_id)+'.png',vote_mask)vote_per_image(3)

模型融合后的预测结果

额外的思路

1、GAN -Image-to-Image Translation with Conditional Adversarial Nets（pix2pix: generate some fake satellite images to enlarge the dataset）

针对数据集小的问题：使用生成对抗网络生成虚假的卫星地图(用标注好的卫星地图生成虚假的卫星地图) - 进一步扩大数据集 - 使用这些虚假+真实的数据集训练网络 - 网络的泛化能力将有更大的提升

问题：由于标注得不好，生成的虚假卫星地图质量不好（如下右图）

2、DeepLab

3、Mask RCNN

4、FCN

5、RefineNet

6、Post-Processing: CRF

基于ResNet+U-Net和Mask-R-CNN的卫星地图建筑物分割

地图图像识别的目标

在卫星图片上标注出建筑物轮廓：该分割目标与其它大型比赛（如微软的COCO Challenge、谷歌的Google AI Open Images比赛）相比，物体类别单一，且图片质量均匀

数据集

如图所示，是一组人工标注完善的卫星图片，其Mask和建筑物匹配度高，该图来自Crowdai上的比赛Mapping Challenge

地图图像识别数据集准备

生成训练样本

训练数据来源：都不是人工直接标注出的卫星图像，而是有一个由人工打上地理标记的图层文件(shapefile格式 - GIS领域的标准数据格式)以及从Google Map上抓取到的对应地区的卫星图片

构建训练数据：将图层文件中的标记(经纬度坐标)映射到Google Map卫星图片上(以图片左上角为原点，向右为X轴，向下为Y轴)

方法：Google Map JavaScript API的转换方式（用Python语言重写实现）

标注数据的格式

生成的数据集都是COCO风格的标注数据（COCO标注数据的具体规范可以参考/cocodataset/cocoapi这个Github repo里的示例代码，在Windows安装Pycocotools的话可以参考/philferriere/cocoapi)

问题：目前生成的训练数据中，标注与真实房屋的位置，很多图像上有大约10-20个像素的偏差(图片大小300 x 300)，也有不少标注大于房屋实际面积的情况。甚至存在标注面积实际为房屋面积的2倍以上，这将导致如果精准分割出房屋，使用IoU>0.5作为阈值过滤掉不合格预测结果，再计算准确率的话，很多实际上完美分割的结果，会被认为是无效的。如图（红色边框较蓝色房屋，大小相似，但是位置偏移；黄色边框将绿色的建筑物全部囊括，但是面积要大很多，此时如果完美分割绿色建筑物，IoU很可能由于小于0.5而无效）：

深度神经网络模型

ResNet+U-Net

Crowdai上举办了Open Map Challenge，其所解决的问题和这个问题相近，排名第一的队伍是Neptume.ml公司（其Github Repo地址：/neptune-ml/open-solution-mapping-challenge）

其所使用的模型是ResNet 101和U-Net的组合，使用预训练的ResNet101对图像进行特征提取，再使用U-Net进行图像分割

模型的损失函数由两部分组成：loss = binary_cross_entropy * weight1 + dice_loss * weight2

其中，Binary Cross Entropy是计算预测值与实际标注每一个像素的异同，Dice Loss是用IoU的思想计算预测值与实际标注的偏差，两种Loss值的权重是需要人为设定的超参数，根据Github中的描述，模型训练前期，需要更多考虑Binary Cross Entropy损失值

Mask-R-CNN

比赛的主办方，给出的Baselline模型是Mask-R-CNN模型（Github：/crowdAI/crowdai-mapping-challenge-mask-rcnn）

相较比赛第一名使用的RestNet+U-Net的方式，Mask-R-CNN模型太重型，这个模型一般用于解决复杂场景下的图像分类、物体检测和语义分割问题

问题：生成的训练集质量较低，导致肉眼评估模型，觉得模型表现尚可，但是使用程序比较预测结果的查全率/查准率(IoU >= 0.5)，结果很差