前言
AI实战：上海垃圾分类系列（一）之快速搭建垃圾分类模型
AI实战：上海垃圾分类系列（二）之快速搭建垃圾分类模型后台服务
AI实战：上海垃圾分类系列（三）之快速搭建垃圾分类智能问答机器人
前一篇文章 AI实战：上海垃圾分类系列之快速搭建垃圾分类模型 ，本文在其之上搭建了一个基于Django、REST的web服务，实现了从浏览器中上传图片做垃圾分类的功能。
2019上海市生活垃圾按照以下标准分类!：http://sh.bendibao.com/zffw/2019225/202535.shtm
上海生活垃圾分类标准及投放要求：https://www.sohu.com/a/163450869_688983
模型介绍
垃圾分类模型详情请查看这里：AI实战：上海垃圾分类系列之快速搭建垃圾分类模型
环境
- Python 3.5
- Django 1.10.8
- Django Rest Framework 3.5
- Pillow
安装环境
pip install -r requirements.txt
垃圾分类识别模型
具体见：./refuse_recognize_service/refuse_recognize/readme.md
配置修改
-
1、修改checkpoints
./refuse_recognize_service/refuse_recognize/runs/checkpoints
把路径改为自己的全路径 -
2、修改html样式：
./refuse_recognize_service/imageupload_frontend/static/index.html
核心代码
models.py :
import os
import uuid
from PIL import Image
from django.db import models
from django.conf import settings
import sys, time
sys.path.append('./refuse_recognize/')
sys.path.append('./refuse_recognize/textcnn/')
from refuse import *
refuse_classification = RefuseRecognize()#加载垃圾识别模型
def scramble_uploaded_filename(instance, filename):
"""
Scramble / uglify the filename of the uploaded file, but keep the files extension (e.g., .jpg or .png)
:param instance:
:param filename:
:return:
"""
extension = filename.split(".")[-1]
return "{}.{}".format(uuid.uuid4(), extension)
def create_thumbnail(input_image, thumbnail_size=(500, 500)):
"""
Create a thumbnail of an existing image
:param input_image:
:param thumbnail_size:
:return:
"""
# make sure an image has been set
if not input_image or input_image == "":
return
# open image
image = Image.open(input_image)
# use PILs thumbnail method; use anti aliasing to make the scaled picture look good
image.thumbnail(thumbnail_size, Image.ANTIALIAS)
# parse the filename and scramble it
filename = scramble_uploaded_filename(None, os.path.basename(input_image.name))
arrdata = filename.split(".")
# extension is in the last element, pop it
extension = arrdata.pop()
basename = "".join(arrdata)
# add _thumb to the filename
new_filename = basename + "_thumb." + extension
# save the image in MEDIA_ROOT and return the filename
image.save(os.path.join(settings.MEDIA_ROOT, new_filename))
return new_filename
class UploadedImage(models.Model):
"""
Provides a Model which contains an uploaded image aswell as a thumbnail
"""
image = models.ImageField("Uploaded image", upload_to=scramble_uploaded_filename)
# thumbnail
thumbnail = models.ImageField("Thumbnail of uploaded image", blank=True)
# title and description
title = models.CharField("Title of the uploaded image", max_length=800, default="Unknown Picture")
description = models.TextField("Description of the uploaded image", default="")
def __str__(self):
return self.title
def save(self, force_insert=False, force_update=False, using=None, update_fields=None):
"""
On save, generate a new thumbnail
:param force_insert:
:param force_update:
:param using:
:param update_fields:
:return:
"""
# generate and set thumbnail or none
self.thumbnail = create_thumbnail(self.image)
# 垃圾类别识别
image_data = Image.open(self.image)
img_path = str(time.time) + '.png'
while os.path.exists(img_path):
img_path = str(time.time) + '.png'
image_data.save(img_path, 'png')
image_data = tf.gfile.FastGFile(img_path, 'rb').read()
classify_res = refuse_classification.recognize_image(image_data)
self.title = classify_res
os.remove(img_path)
# force update as we just changed something
super(UploadedImage, self).save(force_update=force_update)
HTML显示样式
index.html ：
<!DOCTYPE html>
<html lang="en-US">
<head>
<meta charset="utf-8">
<!-- Include Angular and several angular libraries -->
<script src="node_modules/angular/angular.js"></script>
<script src="node_modules/angular-resource/angular-resource.min.js"></script>
<script src="node_modules/ng-file-upload/dist/ng-file-upload.min.js"></script>
<script src="node_modules/angular-animate/angular-animate.min.js"></script>
<script src="node_modules/angularjs-toaster/toaster.js"></script>
<!-- Include our app -->
<script src="js/app.js"></script>
<!-- Include our own controllers, factories, directives, etc... -->
<script src="js/images.rest.js"></script>
<!-- Include Bootstrap CSS -->
<link rel="stylesheet" href="node_modules/bootstrap/dist/css/bootstrap.min.css">
<!-- Include AngularJS Toaster CSS -->
<link rel="stylesheet" href="node_modules/angularjs-toaster/toaster.min.css">
<title>垃圾分类智能识别</title>
</head>
<body>
<div class="container-fluid">
<div ng-app="imageuploadFrontendApp" ng-controller="MainCtrl">
<toaster-container></toaster-container>
<div class="panel panel-default">
<div class="panel-body">
<form class="form" name="form" ng-submit="uploadImage()">
<div class="form-group">
<h3 for="file">选择一张图片：</h3>
<input type="file" ngf-select ng-model="newImage.image" name="file"
class="form-control"
accept="image/*" ngf-max-size="10MB" required
ngf-model-invalid="errorFile">
<i ng-show="form.file.$error.maxSize">File too large
{{ errorFile.size / 1000000 | number:1 }} MB: max 10M</i>
<img ng-show="form.file.$valid" ngf-thumbnail="newImage.image" class="img-responsive" style="max-width: 50%">
<button class="btn btn-warning" ng-click="newImage.image = null" ng-show="newImage.image">Remove</button>
</div>
<button class="btn btn-primary" type="submit">
上传
</button>
</form>
</div>
</div>
<div ng-if="images.length == 0">
请选择一张图片上传！
</div>
<!-- display the images -->
<div class="row">
<div class="col-xs-12 col-sm-12 col-md-6 col-lg-4" ng-repeat="image in images track by image.pk">
<a href="{{ image.image }}">
<img class="img-responsive" ng-src="{{ image.thumbnail }}">
<p>{{ image.description }}</p>
</a>
<h3>垃圾类别：
</h3>
<textarea cols="50" rows="5" id="textareaid">{{ image.title }}</textarea>
<h3>
<button class="btn btn-warning" ng-click="deleteImage(image)">删除</button>
</h3>
</div>
</div>
</div>
</div>
</body>
</html>
开启服务
-
服务
cd refuse_recognize_service python manage.py migrate python manage.py runserver & # 后台运行
效果展示
-
在浏览器中打开 http://localhost:8000/static/index.html （我是在本机测试的，服务器测试的话修改localhost为相应的ip即可）。
-
起始页面
- 上传效果图
- 点击上传后，垃圾分类结果效果图
效果展示大概就是这样了，整体比较粗糙，识别准确率就不再优化了。
还有很多细节工作需要做，主要是垃圾类别标注数据、imagenet的结果的1000类中文对照优化等工作。
完整工程代码
完整的工程包括：
1、完整代码
2、完整的数据
3、完整的垃圾分类识别模型
4、文档
工程下载地址： 快速搭建垃圾分类模型后台服务
扫一扫
2415
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
