java lire es cedd 图片相似搜索

我最喜欢直接贴代码

es 8.2.2


            co.elastic.clients
            elasticsearch-java
            8.2.2
        
        
            org.elasticsearch.client
            elasticsearch-rest-client
            8.2.2
        
        
            org.elasticsearch
            elasticsearch-x-content
            8.2.2
        
        
            com.github.zengde
            lire
            1.0b2
        

    // 获取csclient
    private static ElasticsearchClient createElasticsearchClient() {
        String host = "10.6.30.139";
        Integer port = 9200;
        RestClient restClient = RestClient.builder(new HttpHost(host, port, "http")).build();
        ElasticsearchTransport transport = new RestClientTransport(restClient, new JacksonJsonpMapper());
        ElasticsearchClient client = new ElasticsearchClient(transport);
        System.out.println(client);
        return client;
    }


// 新建索引 dims指定144维度
private static void createIndex() throws IOException {
        String mapping = "{" +
                "  \"mappings\": {" +
                /*                "    \"product\":{" +*/
                "      \"properties\": {" +
                "        \"imageVector\": {" +
                "          \"type\": \"dense_vector\"," +
                "          \"dims\": 144," +
                "          \"index\": true,\n" +
                "          \"similarity\": \"dot_product\"" +
                "        }," +
                "        \"imageUrl\": {" +
                "          \"type\": \"text\"" +
                "        }" +
                "      }" +
                /*"    }" +*/
                "  }" +
                "}";
        // 额外指定 index 为 true 是因为，为了实现 _knn_search，ES 必须在底层构建一个新的数据结构（目前使用的是 HNSW graph ）。
        // similarity 指定向量相似度算法，可以是 l2_norm 、dot_product、cosine
        CreateIndexRequest request = new CreateIndexRequest.Builder().index(INDEX_NAME).withJson(IOUtils.toInputStream(mapping, StandardCharsets.UTF_8)).build();
        CreateIndexResponse response = client.indices().create(request);
        System.out.println(response);
    }

//根据图片地址返回144维度cedd向量
private static double[] imageVector(String url) throws Exception {
        url = OssUtils.processPic(url, "?x-oss-process=image/resize,m_lfit,h_250,w_250");
        System.out.println(url);
        BufferedImage img = ImageIO.read(new URL(url));
        CEDD lireFeature = new CEDD();
        lireFeature.extract(img);
        byte[] byteHistogram = lireFeature.getByteHistogram();
        printArray(byteHistogram);
        double[] ds = SerializationUtils.castToDoubleArray(byteHistogram);
        printArray(ds);
        return ds;
    }

//根据图片地址返回144维度cedd向量
    private static List imageVectorList(String url) throws Exception {
        double[] searchVector = imageVector(url);
        List vectors = new ArrayList<>();
        for (double d : searchVector) {
            vectors.add(d);
        }
        return vectors;
    }

// 增加索引
    public static void index(String url, Long productId) throws Exception {
        Map doc = new HashMap();
        doc.put("imageVector", imageVector(url));
        doc.put("imageUrl", OssUtils.processPic(url, ""));
        IndexRequest