# -*- coding: utf-8 -*- # @Time : 2024/6/4 21:34 # @Author : GG # @File : classifer.py # @Software: PyCharm import os import cv2 import json import utils import joblib import logging import random import numpy as np from PIL import Image from sklearn.ensemble import RandomForestRegressor #图像分类网络所需库，实际并未使用分类网络 # import torch # import torch.nn as nn # from torchvision import transforms #番茄RGB处理模型 class Tomato: def __init__(self): ''' 初始化 Tomato 类。''' pass def extract_s_l(self, image): ''' 提取图像的 S 通道（饱和度）和 L 通道（亮度），并将两者相加。 :param image: 输入的 BGR 图像 :return: S 通道和 L 通道相加的结果 ''' hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) lab = cv2.cvtColor(image, cv2.COLOR_BGR2Lab) s_channel = hsv[:, :, 1] l_channel = lab[:, :, 0] result = cv2.add(s_channel, l_channel) return result def find_reflection(self, image, threshold=190): ''' 通过阈值处理识别图像中的反射区域。 :param image: 输入的单通道图像 :param threshold: 用于二值化的阈值 :return: 二值化后的图像，高于阈值的部分为白色，其余为黑色 ''' _, reflection = cv2.threshold(image, threshold, 255, cv2.THRESH_BINARY) return reflection def otsu_threshold(self, image): ''' 使用 Otsu 大津法自动计算并应用阈值，进行图像的二值化处理。 :param image: 输入的单通道图像 :return: 二值化后的图像 ''' _, binary = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) return binary def extract_g_r(self, image): ''' 提取图像中的 G 通道（绿色），放大并减去 R 通道（红色）。 :param image: 输入的 BGR 图像 :return: G 通道乘以 1.5 后减去 R 通道的结果 ''' g_channel = image[:, :, 1] r_channel = image[:, :, 2] result = cv2.subtract(cv2.multiply(g_channel, 1.5), r_channel) return result def extract_r_b(self, image): ''' 提取图像中的 R 通道（红色）和 B 通道（蓝色），并进行相减。 :param image: 输入的 BGR 图像 :return: R 通道减去 B 通道的结果 ''' r_channel = image[:, :, 2] b_channel = image[:, :, 0] result = cv2.subtract(r_channel, b_channel) return result def extract_r_g(self, image): ''' 提取图像中的 R 通道（红色）和 G 通道（绿色），并进行相减。 :param image: 输入的 BGR 图像 :return: R 通道减去 G 通道的结果 ''' r_channel = image[:, :, 2] g_channel = image[:, :, 1] result = cv2.subtract(r_channel, g_channel) return result def threshold_segmentation(self, image, threshold, color=255): ''' 对图像进行阈值分割，高于阈值的部分设置为指定的颜色。 :param image: 输入的单通道图像 :param threshold: 阈值 :param color: 设置的颜色值 :return: 分割后的二值化图像 ''' _, result = cv2.threshold(image, threshold, color, cv2.THRESH_BINARY) return result def bitwise_operation(self, image1, image2, operation='and'): ''' 对两幅图像执行位运算（与或运算）。 :param image1: 第一幅图像 :param image2: 第二幅图像 :param operation: 执行的操作类型（'and' 或 'or'） :return: 位运算后的结果 ''' if operation == 'and': result = cv2.bitwise_and(image1, image2) elif operation == 'or': result = cv2.bitwise_or(image1, image2) else: raise ValueError("operation must be 'and' or 'or'") return result def largest_connected_component(self, bin_img): ''' 提取二值图像中的最大连通区域。 :param bin_img: 输入的二值图像 :return: 只包含最大连通区域的二值图像 ''' num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(bin_img, connectivity=8) if num_labels <= 1: return np.zeros_like(bin_img) largest_label = 1 + np.argmax(stats[1:, cv2.CC_STAT_AREA]) new_bin_img = np.zeros_like(bin_img) new_bin_img[labels == largest_label] = 255 return new_bin_img def close_operation(self, bin_img, kernel_size=(5, 5)): ''' 对二值图像进行闭运算，用于消除内部小孔和连接接近的对象。 :param bin_img: 输入的二值图像 :param kernel_size: 核的大小 :return: 进行闭运算后的图像 ''' kernel = cv2.getStructuringElement(cv2.MORPH_RECT, kernel_size) closed_img = cv2.morphologyEx(bin_img, cv2.MORPH_CLOSE, kernel) return closed_img def open_operation(self, bin_img, kernel_size=(5, 5)): ''' 对二值图像进行开运算，用于去除小的噪点。 :param bin_img: 输入的二值图像 :param kernel_size: 核的大小 :return: 进行开运算后的图像 ''' kernel = cv2.getStructuringElement(cv2.MORPH_RECT, kernel_size) opened_img = cv2.morphologyEx(bin_img, cv2.MORPH_OPEN, kernel) return opened_img def draw_tomato_edge(self, original_img, bin_img): ''' 在原始图像上绘制最大西红柿轮廓的近似多边形。 :param original_img: 原始 BGR 图像 :param bin_img: 西红柿的二值图像 :return: 带有绘制边缘的原始图像和边缘掩码 ''' bin_img_processed = self.close_operation(bin_img, kernel_size=(15, 15)) contours, _ = cv2.findContours(bin_img_processed, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) if not contours: return original_img, np.zeros_like(bin_img) max_contour = max(contours, key=cv2.contourArea) epsilon = 0.0006 * cv2.arcLength(max_contour, True) approx = cv2.approxPolyDP(max_contour, epsilon, True) cv2.drawContours(original_img, [approx], -1, (0, 255, 0), 3) mask = np.zeros_like(bin_img) cv2.drawContours(mask, [max_contour], -1, (255), thickness=cv2.FILLED) return original_img, mask def draw_tomato_edge_convex_hull(self, original_img, bin_img): ''' 在原始图像上绘制最大西红柿轮廓的凸包。 :param original_img: 原始 BGR 图像 :param bin_img: 西红柿的二值图像 :return: 带有绘制凸包的原始图像 ''' bin_img_blurred = cv2.GaussianBlur(bin_img, (5, 5), 0) contours, _ = cv2.findContours(bin_img_blurred, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) if not contours: return original_img max_contour = max(contours, key=cv2.contourArea) hull = cv2.convexHull(max_contour) cv2.drawContours(original_img, [hull], -1, (0, 255, 0), 3) return original_img def fill_holes(self, bin_img): ''' 使用 floodFill 算法填充图像中的孔洞。 :param bin_img: 输入的二值图像 :return: 填充后的图像 ''' img_filled = bin_img.copy() height, width = bin_img.shape mask = np.zeros((height + 2, width + 2), np.uint8) cv2.floodFill(img_filled, mask, (0, 0), 255) img_filled_inv = cv2.bitwise_not(img_filled) img_filled = cv2.bitwise_or(bin_img, img_filled_inv) return img_filled def bitwise_and_rgb_with_binary(self, rgb_img, bin_img): ''' 将 RGB 图像与二值图像进行按位与操作，用于将二值区域应用于原始图像。 :param rgb_img: 原始 RGB 图像 :param bin_img: 二值图像 :return: 按位与后的结果图像 ''' bin_img_3channel = cv2.cvtColor(bin_img, cv2.COLOR_GRAY2BGR) result = cv2.bitwise_and(rgb_img, bin_img_3channel) return result def extract_max_connected_area(self, image, lower_hsv, upper_hsv): ''' 提取图像中满足 HSV 范围条件的最大连通区域，并填充孔洞。 :param image: 输入的 BGR 图像 :param lower_hsv: HSV 范围的下限 :param upper_hsv: HSV 范围的上限 :return: 处理后的图像 ''' hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV) mask = cv2.inRange(hsv, lower_hsv, upper_hsv) num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(mask, connectivity=8) largest_label = 1 + np.argmax(stats[1:, cv2.CC_STAT_AREA]) new_bin_img = np.zeros_like(mask) new_bin_img[labels == largest_label] = 255 img_filled = new_bin_img.copy() height, width = new_bin_img.shape mask = np.zeros((height + 2, width + 2), np.uint8) cv2.floodFill(img_filled, mask, (0, 0), 255) img_filled_inv = cv2.bitwise_not(img_filled) img_filled = cv2.bitwise_or(new_bin_img, img_filled_inv) return img_filled #百香果RGB处理模型 class Passion_fruit: def __init__(self, hue_value=37, hue_delta=10, value_target=25, value_delta=10): # 初始化常用参数 self.hue_value = hue_value self.hue_delta = hue_delta self.value_target = value_target self.value_delta = value_delta def create_mask(self, hsv_image): # 创建H通道阈值掩码 lower_hue = np.array([self.hue_value - self.hue_delta, 0, 0]) upper_hue = np.array([self.hue_value + self.hue_delta, 255, 255]) hue_mask = cv2.inRange(hsv_image, lower_hue, upper_hue) # 创建V通道排除中心值的掩码 lower_value_1 = np.array([0, 0, 0]) upper_value_1 = np.array([180, 255, self.value_target - self.value_delta]) lower_value_2 = np.array([0, 0, self.value_target + self.value_delta]) upper_value_2 = np.array([180, 255, 255]) value_mask_1 = cv2.inRange(hsv_image, lower_value_1, upper_value_1) value_mask_1 = cv2.bitwise_not(value_mask_1) value_mask_2 = cv2.inRange(hsv_image, lower_value_2, upper_value_2) value_mask = cv2.bitwise_and(value_mask_1, value_mask_2) # 合并H通道和V通道掩码 return cv2.bitwise_and(hue_mask, value_mask) def apply_morphology(self, mask): # 应用形态学操作 kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5)) return cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel) def find_largest_component(self, mask): # 寻找最大连通组件 num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(mask, 4, cv2.CV_32S) if num_labels < 2: return None # 没有找到显著的组件 max_label = 1 + np.argmax(stats[1:, cv2.CC_STAT_AREA]) # 跳过背景 return (labels == max_label).astype(np.uint8) * 255 def draw_contours_on_image(self, original_image, mask_image): """ 在原图上绘制轮廓 :param original_image: 原图的NumPy数组 :param mask_image: 轮廓mask的NumPy数组 :return: 在原图上绘制轮廓后的图像 """ # 确保mask_image是二值图像 _, binary_mask = cv2.threshold(mask_image, 127, 255, cv2.THRESH_BINARY) # 查找mask图像中的轮廓 contours, _ = cv2.findContours(binary_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 在原图上绘制轮廓 cv2.drawContours(original_image, contours, -1, (0, 255, 0), 2) return original_image def bitwise_and_rgb_with_binary(self, rgb_img, bin_img): ''' 将 RGB 图像与二值图像进行按位与操作，用于将二值区域应用于原始图像。 :param rgb_img: 原始 RGB 图像 :param bin_img: 二值图像 :return: 按位与后的结果图像 ''' # 检查 RGB 图像是否为空或全黑 if rgb_img is None or rgb_img.size == 0 or np.all(rgb_img == 0): logging.warning("RGB 图像为空或全黑，返回一个全黑RGB图像。") return np.zeros((100, 100, 3), dtype=np.uint8) if rgb_img is None else np.zeros_like(rgb_img) # 检查二值图像是否为空或全黑 if bin_img is None or bin_img.size == 0 or np.all(bin_img == 0): logging.warning("二值图像为空或全黑，返回一个全黑RGB图像。") return np.zeros((100, 100, 3), dtype=np.uint8) if rgb_img is None else np.zeros_like(rgb_img) # 转换二值图像为三通道 try: bin_img_3channel = cv2.cvtColor(bin_img, cv2.COLOR_GRAY2BGR) except cv2.error as e: logging.error(f"转换二值图像时发生错误: {e}") return np.zeros_like(rgb_img) # 进行按位与操作 try: result = cv2.bitwise_and(rgb_img, bin_img_3channel) except cv2.error as e: logging.error(f"执行按位与操作时发生错误: {e}") return np.zeros_like(rgb_img) return result #糖度预测模型 class Spec_predict(object): def __init__(self, load_from=None, debug_mode=False): self.debug_mode = debug_mode self.log = utils.Logger(is_to_file=debug_mode) if load_from is not None: self.load(load_from) else: self.model = RandomForestRegressor(n_estimators=100) def load(self, path): if not os.path.isabs(path): self.log.log('Path is relative, converting to absolute path.') path = os.path.abspath(path) if not os.path.exists(path): self.log.log(f'Model file not found at path: {path}') raise FileNotFoundError(f'Model file not found at path: {path}') with open(path, 'rb') as f: model_dic = joblib.load(f) self.model = model_dic self.log.log(f'Model loaded successfully from {path}') def predict(self, data_x): ''' 对数据进行预测 :param data_x: 波段选择后的数据 :return: 预测结果二值化后的数据，0为背景，1为黄芪,2为杂质2，3为杂质1，4为甘草片，5为红芪 ''' data_x = data_x.reshape(1, -1) selected_bands = [8, 9, 10, 48, 49, 50, 77, 80, 103, 108, 115, 143, 145] data_x = data_x[:, selected_bands] data_y = self.model.predict(data_x) return data_y[0] #数据处理模型 class Data_processing: def __init__(self): pass def contour_process(self, image_array): # 检查图像是否为空或全黑 if image_array is None or image_array.size == 0 or np.all(image_array == 0): logging.warning("输入的图像为空或全黑，返回一个全黑图像。") return np.zeros_like(image_array) if image_array is not None else np.zeros((100, 100), dtype=np.uint8) # 应用中值滤波 image_filtered = cv2.medianBlur(image_array, 5) # 形态学闭操作 kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (15, 15)) image_closed = cv2.morphologyEx(image_filtered, cv2.MORPH_CLOSE, kernel) # 查找轮廓 contours, _ = cv2.findContours(image_closed, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 创建空白图像以绘制轮廓 image_contours = np.zeros_like(image_array) # 进行多边形拟合并填充轮廓 for contour in contours: epsilon = 0.001 * cv2.arcLength(contour, True) approx = cv2.approxPolyDP(contour, epsilon, True) if cv2.contourArea(approx) > 100: # 仅处理较大的轮廓 cv2.drawContours(image_contours, [approx], -1, (255, 255, 255), -1) return image_contours def analyze_ellipse(self, image_array): # 查找白色区域的轮廓 _, binary_image = cv2.threshold(image_array, 127, 255, cv2.THRESH_BINARY) contours, _ = cv2.findContours(binary_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 初始化变量用于存储最大轮廓的长径和短径 major_axis = 0 minor_axis = 0 # 对每个找到的轮廓，找出可以包围它的最小椭圆，并计算长径和短径 for contour in contours: if len(contour) >= 5: # 至少需要5个点来拟合椭圆 ellipse = cv2.fitEllipse(contour) (center, axes, orientation) = ellipse major_axis0 = max(axes) minor_axis0 = min(axes) # 更新最大的长径和短径 if major_axis0 > major_axis: major_axis = major_axis0 minor_axis = minor_axis0 return major_axis, minor_axis def analyze_defect(self, image_array): # 查找白色区域的轮廓 _, binary_image = cv2.threshold(image_array, 127, 255, cv2.THRESH_BINARY) contours_white, _ = cv2.findContours(binary_image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) # 初始化统计数据 count_black_areas = 0 total_pixels_black_areas = 0 s = 0.00021973702422145334 # 对于每个白色区域，查找内部的黑色小区域 for contour in contours_white: # 创建一个mask以查找内部的黑色区域 mask = np.zeros_like(image_array) cv2.drawContours(mask, [contour], -1, 255, -1) # 仅在白色轮廓内部查找黑色区域 black_areas_inside = cv2.bitwise_and(cv2.bitwise_not(image_array), mask) # 查找黑色区域的轮廓 contours_black, _ = cv2.findContours(black_areas_inside, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) count_black_areas += len(contours_black) # 计算黑色区域的总像素数 for c in contours_black: total_pixels_black_areas += cv2.contourArea(c) number_defects = count_black_areas total_pixels = total_pixels_black_areas * s return number_defects, total_pixels def weight_estimates(self, long_axis, short_axis): """ 根据西红柿的长径、短径和直径估算其体积。 使用椭圆体积公式计算体积。 参数: diameter (float): 西红柿的直径 long_axis (float): 西红柿的长径 short_axis (float): 西红柿的短径 返回: float: 估算的西红柿体积 """ density = 0.652228972 a = ((long_axis / 425) * 6.3) / 2 b = ((short_axis / 425) * 6.3) / 2 volume = 4 / 3 * np.pi * a * b * b weight = round(volume * density) #重量单位为g return weight def analyze_tomato(self, img): """ 分析给定图像，提取和返回西红柿的长径、短径、缺陷数量和缺陷总面积，并返回处理后的图像。 使用 Tomoto 类的图像处理方法，以及自定义的尺寸和缺陷信息获取函数。 参数: img (numpy.ndarray): 输入的 BGR 图像 返回: tuple: (长径, 短径, 缺陷区域个数, 缺陷区域总像素, 处理后的图像) """ tomato = Tomato() # 创建 Tomato 类的实例 # 设置 S-L 通道阈值并处理图像 threshold_s_l = 180 threshold_fore_g_r_t = 20 img = cv2.cvtColor(img,cv2.COLOR_RGB2BGR) s_l = tomato.extract_s_l(img) thresholded_s_l = tomato.threshold_segmentation(s_l, threshold_s_l) new_bin_img = tomato.largest_connected_component(thresholded_s_l) # 绘制西红柿边缘并获取缺陷信息 edge, mask = tomato.draw_tomato_edge(img, new_bin_img) org_defect = tomato.bitwise_and_rgb_with_binary(edge, new_bin_img) fore = tomato.bitwise_and_rgb_with_binary(img, mask) fore_g_r_t = tomato.threshold_segmentation(tomato.extract_g_r(fore), threshold=threshold_fore_g_r_t) res = cv2.bitwise_or(new_bin_img, fore_g_r_t) nogreen = tomato.bitwise_and_rgb_with_binary(edge, res) # 统计白色像素点个数 # print(np.sum(fore_g_r_t == 255)) # print(np.sum(mask == 255)) # print(np.sum(fore_g_r_t == 255) / np.sum(mask == 255)) green_percentage = np.sum(fore_g_r_t == 255) / np.sum(mask == 255) green_percentage = round(green_percentage, 2) # 获取西红柿的尺寸信息 long_axis, short_axis = self.analyze_ellipse(mask) # 获取缺陷信息 number_defects, total_pixels = self.analyze_defect(new_bin_img) # 将处理后的图像转换为 RGB 格式 rp = cv2.cvtColor(nogreen, cv2.COLOR_BGR2RGB) #直径单位为cm，所以需要除以10 diameter = (long_axis + short_axis) /425 * 63 / 2 / 10 # print(f'直径：{diameter}') # 如果直径小于3，判断为空果拖异常图，则将所有值重置为0 if diameter < 2.5: diameter = 0 green_percentage = 0 number_defects = 0 total_pixels = 0 rp = cv2.cvtColor(np.ones((613, 800, 3), dtype=np.uint8), cv2.COLOR_BGR2RGB) return diameter, green_percentage, number_defects, total_pixels, rp return diameter, green_percentage, number_defects, total_pixels, rp def analyze_passion_fruit(self, img, hue_value=37, hue_delta=10, value_target=25, value_delta=10): if img is None: logging.error("Error: 无图像数据.") return None # 创建PassionFruit类的实例 pf = Passion_fruit(hue_value=hue_value, hue_delta=hue_delta, value_target=value_target, value_delta=value_delta) img = cv2.cvtColor(img,cv2.COLOR_RGB2BGR) hsv_image = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) combined_mask = pf.create_mask(hsv_image) combined_mask = pf.apply_morphology(combined_mask) max_mask = pf.find_largest_component(combined_mask) contour_mask = self.contour_process(max_mask) long_axis, short_axis = self.analyze_ellipse(contour_mask) #重量单位为g，加上了一点随机数 weight = self.weight_estimates(long_axis, short_axis) weight = (weight * 2) + random.randint(0, 30) - random.randint(0, 30) number_defects, total_pixels = self.analyze_defect(max_mask) edge = pf.draw_contours_on_image(img, contour_mask) org_defect = pf.bitwise_and_rgb_with_binary(edge, max_mask) rp = cv2.cvtColor(org_defect, cv2.COLOR_BGR2RGB) #直径单位为cm，所以需要除以10 diameter = (long_axis + short_axis) /425 * 63 / 2 / 10 # print(f'直径：{diameter}') return diameter, weight, number_defects, total_pixels, rp # #下面封装的是ResNet18和ResNet34的网络模型构建 # #原定用于构建RGB图像有果无果判断，后续发现存在纰漏，暂时搁置并未实际使用 # class BasicBlock(nn.Module): # ''' # BasicBlock for ResNet18 and ResNet34 # # ''' # expansion = 1 # # def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs): # super(BasicBlock, self).__init__() # self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel, # kernel_size=3, stride=stride, padding=1, bias=False) # self.bn1 = nn.BatchNorm2d(out_channel) # self.relu = nn.ReLU() # self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel, # kernel_size=3, stride=1, padding=1, bias=False) # self.bn2 = nn.BatchNorm2d(out_channel) # self.downsample = downsample # # def forward(self, x): # identity = x # if self.downsample is not None: # identity = self.downsample(x) # # out = self.conv1(x) # out = self.bn1(out) # out = self.relu(out) # # out = self.conv2(out) # out = self.bn2(out) # # out += identity # out = self.relu(out) # # return out # # class ResNet(nn.Module): # ''' # ResNet18 and ResNet34 # ''' # def __init__(self, # block, # blocks_num, # num_classes=1000, # include_top=True, # groups=1, # width_per_group=64): # super(ResNet, self).__init__() # self.include_top = include_top # self.in_channel = 64 # # self.groups = groups # self.width_per_group = width_per_group # # self.conv1 = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2, # padding=3, bias=False) # self.bn1 = nn.BatchNorm2d(self.in_channel) # self.relu = nn.ReLU(inplace=True) # self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) # self.layer1 = self._make_layer(block, 64, blocks_num[0]) # self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2) # self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2) # self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2) # if self.include_top: # self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # output size = (1, 1) # self.fc = nn.Linear(512 * block.expansion, num_classes) # # for m in self.modules(): # if isinstance(m, nn.Conv2d): # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') # # def _make_layer(self, block, channel, block_num, stride=1): # downsample = None # if stride != 1 or self.in_channel != channel * block.expansion: # downsample = nn.Sequential( # nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False), # nn.BatchNorm2d(channel * block.expansion)) # # layers = [] # layers.append(block(self.in_channel, # channel, # downsample=downsample, # stride=stride, # groups=self.groups, # width_per_group=self.width_per_group)) # self.in_channel = channel * block.expansion # # for _ in range(1, block_num): # layers.append(block(self.in_channel, # channel, # groups=self.groups, # width_per_group=self.width_per_group)) # # return nn.Sequential(*layers) # # def forward(self, x): # x = self.conv1(x) # x = self.bn1(x) # x = self.relu(x) # x = self.maxpool(x) # # x = self.layer1(x) # x = self.layer2(x) # x = self.layer3(x) # x = self.layer4(x) # # if self.include_top: # x = self.avgpool(x) # x = torch.flatten(x, 1) # x = self.fc(x) # # return x # # def resnet18(num_classes=1000, include_top=True): # return ResNet(BasicBlock, [2, 2, 2, 2], num_classes=num_classes, include_top=include_top) # # def resnet34(num_classes=1000, include_top=True): # return ResNet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes, include_top=include_top) # # #图像有无果判别模型 # class ImageClassifier: # ''' # 图像分类器，用于加载预训练的 ResNet 模型并进行图像分类。 # ''' # def __init__(self, model_path, class_indices_path, device=None): # if device is None: # self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # else: # self.device = device # # # 加载类别索引 # assert os.path.exists(class_indices_path), f"File: '{class_indices_path}' does not exist." # with open(class_indices_path, "r") as json_file: # self.class_indict = json.load(json_file) # # # 创建模型并加载权重 # self.model = resnet34(num_classes=len(self.class_indict)).to(self.device) # assert os. path.exists(model_path), f"File: '{model_path}' does not exist." # self.model.load_state_dict(torch.load(model_path, map_location=self.device)) # self.model.eval() # # # 设置图像转换 # self.transform = transforms.Compose([ # transforms.Resize(256), # transforms.CenterCrop(224), # transforms.ToTensor(), # transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # ]) # # def predict(self, image_np): # ''' # 对图像进行分类预测。 # :param image_np: # :return: # ''' # # 将numpy数组转换为图像 # image = Image.fromarray(image_np.astype('uint8'), 'RGB') # image = self.transform(image).unsqueeze(0).to(self.device) # # with torch.no_grad(): # output = self.model(image).cpu() # predict = torch.softmax(output, dim=1) # predict_cla = torch.argmax(predict, dim=1).numpy() # # # return self.class_indict[str(predict_cla[0])] # return predict_cla[0]