深度学习预测中的裁剪拼接去除接缝解决方案（遥感影像）

遥感影像通常是大幅的，因此直接输入predict模型，会out of memory，通常的处理方案是：

先剪裁-预测小幅影像-拼接

一、剪裁与拼接函数实现

以示例影像为例：4240*3120剪裁成256*256：

#剪裁函数

#剪裁函数
def crop_image(image_path, output_path,crop_size=(256, 256), background_color=0):
    image = tf.imread(image_path)
    height, width, bands = image.shape
    new_width = (width // crop_size[0] + 1) * crop_size[0]
    new_height = (height // crop_size[1] + 1) * crop_size[1]
    new_image = np.zeros((new_height, new_width, bands))
    new_image[:height, :width, :] = image

    cropped_images = []
    for i in range(0, new_height, crop_size[1]):
        for j in range(0, new_width, crop_size[0]):
            cropped_image = new_image[i:i +crop_size[1], j:j + crop_size[0], :]
            cropped_images.append(cropped_image)
            tf.imwrite(os.path.join(output_path, f"cropped_{i}_{j}.tif"),cropped_image)
    ret