diff --git a/alg/qrtool.cpp b/alg/qrtool.cpp
index 3e51845..093dfd5 100644
--- a/alg/qrtool.cpp
+++ b/alg/qrtool.cpp
@@ -14,6 +14,8 @@
 #include <stdlib.h>
 #include <unistd.h>
 #include "opencv2/objdetect.hpp"
+#include <opencv2/dnn.hpp>
+
 #include "mq_worker.h"
 #if ENABLE_GRPC
 #include "fileprocess.h"
@@ -607,6 +609,31 @@ void usage(const char *name, vector<string> &cmds)
     }
 }
 
+static
+int roi_verify_cmd(char **argv, int argc)
+{
+    char *model_path = argv[0];
+    char *input_file = argv[1];
+    Mat input_img = imread(input_file);
+    cv::dnn::Net net = cv::dnn::readNetFromONNX(model_path);
+    if (net.empty()) {
+        std::cerr << "Failed to load ONNX model!" << std::endl;
+        return -1;
+    }
+    cv::Mat blob;
+    cv::resize(input_img, input_img, cv::Size(128, 64));  // 调整图像大小
+    cv::dnn::blobFromImage(input_img, blob, 1.0 / 255.0, cv::Size(64, 128), cv::Scalar(0.485, 0.456, 0.406), true, false);
+    blob = (blob - cv::Scalar(0.485, 0.456, 0.406)) / cv::Scalar(0.229, 0.224, 0.225);  // 归一化
+    net.setInput(blob);
+    cv::Mat output = net.forward();
+    float* data = (float*)output.data;
+    int class_id = std::max_element(data, data + output.total()) - data;  // 找到最大概率的类别
+    float confidence = data[class_id];
+    std::cout << "Predicted class: " << class_id << ", Confidence: " << confidence << std::endl;
+
+    return 0;
+}
+
 #ifdef QRTOOL_MAIN
 int main(int argc, char *argv[])
 {
@@ -640,6 +667,7 @@ int main(int argc, char *argv[])
 #endif
     add_cmd(http_server, 1);
     add_cmd(verify, 2);
+    add_cmd(roi_verify, 2);
     usage(argv[0], cmds);
     return 1;
 }