feat: 集成阿里百炼API实现图片内容识别功能

添加DashscopeService接口及实现，用于调用阿里百炼API进行图片内容识别修改PdfImageExtractService增加基于百炼API的图片处理逻辑新增OSS服务方法支持临时文件处理和删除更新配置文件添加百炼模型相关配置
2026-04-08 09:17:33 +00:00 · 2025-06-04 17:55:47 +08:00
parent 53e3180658
commit 1d51a103d0
13 changed files with 472 additions and 83 deletions
--- a/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/controller/knowledge/KnowledgeController.java
+++ b/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/controller/knowledge/KnowledgeController.java
@@ -52,7 +52,7 @@ public class KnowledgeController extends BaseController {

  private final IKnowledgeFragmentService fragmentService;

-//  private final PdfImageExtractService pdfImageExtractService;
+  private final PdfImageExtractService pdfImageExtractService;

  /**
   * 根据用户信息查询本地知识库
@@ -170,11 +170,11 @@ public class KnowledgeController extends BaseController {
   * @param file PDF文件
   * @return 文件名称和图片内容
   */
-//  @PostMapping("/extract-images")
-//  @Operation(summary = "提取PDF中的图片并调用大模型,识别图片内容并返回", description = "提取PDF中的图片并调用gpt-4o-mini,识别图片内容并返回")
-//  public R<List<PdfFileContentResult>> extractImages(
-//      @RequestPart("file") MultipartFile file
-//  ) throws IOException {
-//    return R.ok(pdfImageExtractService.extractImages(file));
-//  }
+  @PostMapping("/extract-images")
+  @Operation(summary = "提取PDF中的图片并调用大模型,识别图片内容并返回", description = "提取PDF中的图片并调用gpt-4o-mini,识别图片内容并返回")
+  public R<List<PdfFileContentResult>> extractImages(
+  ) throws IOException {
+    return R.ok(pdfImageExtractService
+        .dealFileContent4Dashscope("https://hnzuoran02-1327573163.cos.ap-nanjing.myqcloud.com/crmebimage/public/content/2025/06/04/e115264eb22f423ea0b211709361c29f071avy39ez.jpg"));
+  }
 }
--- a/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/service/knowledge/DealFileService.java
+++ b/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/service/knowledge/DealFileService.java
@@ -26,6 +26,7 @@ import org.ruoyi.mapper.KnowledgeAttachPicMapper;
 import org.ruoyi.mapper.KnowledgeFragmentMapper;
 import org.ruoyi.mapper.KnowledgeInfoMapper;
 import org.ruoyi.service.IChatModelService;
+import org.ruoyi.service.PdfImageExtractService;
 import org.ruoyi.service.VectorStoreService;
 import org.ruoyi.service.impl.PdfImageExtractServiceImpl;
 import org.ruoyi.system.domain.vo.SysOssVo;
@@ -64,18 +65,10 @@ public class DealFileService {

  private final ISysOssService ossService;

-//  private final PdfImageExtractService pdfImageExtractService;
+  private final PdfImageExtractService pdfImageExtractService;

  private final KnowledgeAttachPicMapper picMapper;

-  @Value("${pdf.extract.service.url}")
-  private String serviceUrl;
-  @Value("${pdf.extract.ai-api.url}")
-  private String aiApiUrl;
-  @Value("${pdf.extract.ai-api.key}")
-  private String aiApiKey;
-
-
  @Async
  public void dealVectorStatus(KnowledgeAttach attachItem) throws Exception {
    try {
@@ -169,8 +162,6 @@ public class DealFileService {
      //获取oss文件
      MultipartFile multipartFile = ossService.downloadByFile(attachItem.getOssId());
      //拆解出图片ZIP
-      PdfImageExtractServiceImpl pdfImageExtractService = new PdfImageExtractServiceImpl(serviceUrl,
-          aiApiUrl, aiApiKey);
      byte[] pngs = pdfImageExtractService.extractImages(multipartFile, "png", true);
      //解压zip，得到图片文件
      MultipartFile[] multipartFiles = ZipUtils.unzipToMultipartFiles(pngs);
@@ -236,6 +227,7 @@ public class DealFileService {

  @Async
  public void dealPicAnysStatus(KnowledgeAttachPic picItem) throws Exception {
+    String filePath = null;
    try {
      //锁定数据 更改 getPicAnysStatus 到进行中
      if (picMapper.update(new LambdaUpdateWrapper<KnowledgeAttachPic>()
@@ -247,13 +239,10 @@ public class DealFileService {
      }
      SysOssVo ossVo = ossService.getById(picItem.getOssId());
      if (ObjectUtil.isNotEmpty(ossVo)) {
-        String fileStr = ossService.downloadByByte(picItem.getOssId());
+        filePath = ossService.downloadToTempPath(picItem.getOssId());
        //调用第三方 分析图片内容
-        PdfImageExtractServiceImpl pdfImageExtractService = new PdfImageExtractServiceImpl(
-            serviceUrl,
-            aiApiUrl, aiApiKey);
-        List<PdfFileContentResult> pdfFileContentResults = pdfImageExtractService.dealFileContent(
-            new String[]{fileStr});
+        List<PdfFileContentResult> pdfFileContentResults = pdfImageExtractService.dealFileContent4DashscopeBase64(
+          filePath);
        if (ObjectUtil.isNotEmpty(pdfFileContentResults)) {
          for (PdfFileContentResult resultItem : pdfFileContentResults) {
            //图片解析内容回写到pic表
@@ -302,6 +291,11 @@ public class DealFileService {
          .eq(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_20)
          .eq(KnowledgeAttachPic::getId, picItem.getId()));
      throw new RuntimeException(e);
+    } finally {
+      //无论成功还是失败，都要删除临时文件
+      if (ObjectUtil.isNotEmpty(filePath)) {
+        ossService.deleteFile(filePath);
+      }
    }
  }

@@ -349,35 +343,32 @@ public class DealFileService {
    return null;
  }

-  public static String parseContent(String jsonString) {
+  public static String parseContent(String content) {
    try {
-      // 创建ObjectMapper实例
+      // 首先尝试作为JSON解析
      ObjectMapper objectMapper = new ObjectMapper();
+      JsonNode rootNode = objectMapper.readTree(content);

-      // 解析JSON字符串
-      JsonNode rootNode = objectMapper.readTree(jsonString);
-
-      // 获取choices数组的第一个元素
+      // 如果是JSON格式，按原有逻辑处理
      JsonNode choicesNode = rootNode.get("choices");
      if (choicesNode != null && choicesNode.isArray() && choicesNode.size() > 0) {
-        // 获取第一个choice
        JsonNode firstChoice = choicesNode.get(0);
-
-        // 获取message节点
        JsonNode messageNode = firstChoice.get("message");
        if (messageNode != null) {
-          // 获取content字段的值
          JsonNode contentNode = messageNode.get("content");
          if (contentNode != null) {
            return contentNode.asText();
          }
        }
+        return "无法找到content内容";
      }
-
-      return "无法找到content内容";
+      
+      // 如果不是预期的JSON格式，直接返回原始内容
+      return content;
+      
    } catch (Exception e) {
-      e.printStackTrace();
-      return "解析JSON时发生错误: " + e.getMessage();
+      // 如果解析JSON失败，说明是普通文本，直接返回
+      return content;
    }
  }

--- a/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/service/knowledge/KnowledgeInfoServiceImpl.java
+++ b/ruoyi-modules/ruoyi-chat/src/main/java/org/ruoyi/chat/service/knowledge/KnowledgeInfoServiceImpl.java
@@ -57,7 +57,6 @@ import org.ruoyi.system.service.ISysOssService;
 import java.io.IOException;
 import java.util.*;

-
 /**
 * 知识库Service业务层处理
 *
@@ -83,7 +82,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {

  private final ISysOssService ossService;

-//  private final PdfImageExtractService pdfImageExtractService;
+  // private final PdfImageExtractService pdfImageExtractService;

  private final KnowledgeAttachPicMapper picMapper;

@@ -170,7 +169,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
   * 保存前的数据校验
   */
  private void validEntityBeforeSave(KnowledgeInfo entity) {
-    //TODO 做一些数据校验,如唯一约束
+    // TODO 做一些数据校验,如唯一约束
  }

  /**
@@ -179,7 +178,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
  @Override
  public Boolean deleteWithValidByIds(Collection<Long> ids, Boolean isValid) {
    if (isValid) {
-      //TODO 做一些业务上的校验,判断是否需要校验
+      // TODO 做一些业务上的校验,判断是否需要校验
    }
    return baseMapper.deleteBatchIds(ids) > 0;
  }
@@ -223,10 +222,10 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
      Collection<Long> ossIds = knowledgeAttachVos.stream()
          .map(KnowledgeAttachVo::getOssId)
          .collect(Collectors.toList());
-      //删除oss
+      // 删除oss
      ossService.deleteWithValidByIds(ossIds, false);

-      //删除图片oss
+      // 删除图片oss
      List<KnowledgeAttachPic> knowledgeAttachPics = picMapper.selectList(
          new LambdaQueryWrapper<KnowledgeAttachPic>()
              .in(KnowledgeAttachPic::getKid,
@@ -234,8 +233,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
                      .collect(Collectors.toList()))
              .in(KnowledgeAttachPic::getAid,
                  knowledgeAttachVos.stream().map(KnowledgeAttachVo::getId)
-                      .collect(Collectors.toList()))
-      );
+                      .collect(Collectors.toList())));
      if (ObjectUtil.isNotEmpty(knowledgeAttachPics)) {
        Collection<Long> tossIds = knowledgeAttachPics.stream()
            .map(KnowledgeAttachPic::getOssId)
@@ -302,7 +300,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
    knowledgeAttach.setCreateTime(new Date());
    if (ObjectUtil.isNotEmpty(uploadDto) && ObjectUtil.isNotEmpty(uploadDto.getOssId())) {
      knowledgeAttach.setOssId(uploadDto.getOssId());
-      //只有pdf文件 才需要拆解图片和分析图片内容
+      // 只有pdf文件 才需要拆解图片和分析图片内容
      if (FileType.PDF.equals(knowledgeAttach.getDocType())) {
        knowledgeAttach.setPicStatus(DealStatus.STATUS_10);
        knowledgeAttach.setPicAnysStatus(DealStatus.STATUS_10);
@@ -310,7 +308,7 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
        knowledgeAttach.setPicStatus(DealStatus.STATUS_30);
        knowledgeAttach.setPicAnysStatus(DealStatus.STATUS_30);
      }
-      //所有文件上传后，都需要同步到向量数据库
+      // 所有文件上传后，都需要同步到向量数据库
      knowledgeAttach.setVectorStatus(DealStatus.STATUS_10);
    }
    attachMapper.insert(knowledgeAttach);
@@ -334,15 +332,14 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
  /**
   * 第一步 定时 拆解PDF文件中的图片
   */
-  //@Scheduled(fixedDelay = 15000) // 每3秒执行一次
+  @Scheduled(fixedDelay = 15000) // 每3秒执行一次
  public void dealKnowledgeAttachPic() throws Exception {
-    //处理 拆解PDF文件中的图片的记录
+    // 处理 拆解PDF文件中的图片的记录
    List<KnowledgeAttach> knowledgeAttaches = attachMapper.selectList(
        new LambdaQueryWrapper<KnowledgeAttach>()
            .eq(KnowledgeAttach::getPicStatus, DealStatus.STATUS_10)
            .eq(KnowledgeAttach::getPicAnysStatus, DealStatus.STATUS_10)
-            .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10)
-    );
+            .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10));
    log.info("===============拆解PDF文件中的图片 size = {}", knowledgeAttaches.size());
    if (ObjectUtil.isNotEmpty(knowledgeAttaches)) {
      for (KnowledgeAttach attachItem : knowledgeAttaches) {
@@ -356,30 +353,29 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
   */
  @Scheduled(fixedDelay = 15000)
  public void dealKnowledgeAttachPicAnys() throws Exception {
-    //获取未处理的图片记录
+    // 获取未处理的图片记录
    List<KnowledgeAttachPic> knowledgeAttachPics = picMapper.selectList(
        new LambdaQueryWrapper<KnowledgeAttachPic>()
            .eq(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_10)
-            .last("LIMIT 20")
-    );
+            .last("LIMIT 20"));
    if (ObjectUtil.isNotEmpty(knowledgeAttachPics)) {
      for (KnowledgeAttachPic picItem : knowledgeAttachPics) {
        dealFileService.dealPicAnysStatus(picItem);
      }
    }
  }
+
  /**
   * 第三步 定时 处理 附件上传后上传向量数据库
   */
  @Scheduled(fixedDelay = 30000) // 每3秒执行一次
  public void dealKnowledgeAttachVector() throws Exception {
-    //处理 需要上传向量数据库的记录
+    // 处理 需要上传向量数据库的记录
    List<KnowledgeAttach> knowledgeAttaches = attachMapper.selectList(
        new LambdaQueryWrapper<KnowledgeAttach>()
            .eq(KnowledgeAttach::getPicStatus, DealStatus.STATUS_30)
            .eq(KnowledgeAttach::getPicAnysStatus, DealStatus.STATUS_30)
-            .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10)
-    );
+            .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10));
    log.info("===============上传向量数据库 size = {}", knowledgeAttaches.size());
    if (ObjectUtil.isNotEmpty(knowledgeAttaches)) {
      for (KnowledgeAttach attachItem : knowledgeAttaches) {
@@ -387,23 +383,24 @@ public class KnowledgeInfoServiceImpl implements IKnowledgeInfoService {
      }
    }
  }
+
  /**
   * 第四步 定时 处理 失败数据
   */
  @Scheduled(fixedDelay = 30 * 60 * 1000)
  public void dealKnowledge40Status() throws Exception {
-      //拆解PDF失败 重新设置状态
-      attachMapper.update(new LambdaUpdateWrapper<KnowledgeAttach>()
-          .set(KnowledgeAttach::getPicStatus, DealStatus.STATUS_10)
-          .eq(KnowledgeAttach::getPicStatus, DealStatus.STATUS_40));
-      //将图片分析失败的数据 重新设置状态
-      picMapper.update(new LambdaUpdateWrapper<KnowledgeAttachPic>()
-          .set(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_10)
-          .eq(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_40));
-      //上传向量库失败 重新设置状态
-      attachMapper.update(new LambdaUpdateWrapper<KnowledgeAttach>()
-          .set(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10)
-          .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_40));
+    // 拆解PDF失败 重新设置状态
+    attachMapper.update(new LambdaUpdateWrapper<KnowledgeAttach>()
+        .set(KnowledgeAttach::getPicStatus, DealStatus.STATUS_10)
+        .eq(KnowledgeAttach::getPicStatus, DealStatus.STATUS_40));
+    // 将图片分析失败的数据 重新设置状态
+    picMapper.update(new LambdaUpdateWrapper<KnowledgeAttachPic>()
+        .set(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_10)
+        .eq(KnowledgeAttachPic::getPicAnysStatus, DealStatus.STATUS_40));
+    // 上传向量库失败 重新设置状态
+    attachMapper.update(new LambdaUpdateWrapper<KnowledgeAttach>()
+        .set(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_10)
+        .eq(KnowledgeAttach::getVectorStatus, DealStatus.STATUS_40));
  }

 }