diff --git a/.github/workflows/release.yaml b/.github/workflows/release.yaml
index 603511b..7c5df52 100644
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -13,17 +13,18 @@ jobs:
     - uses: actions/checkout@v3
       with:
         path: fisher
-    - name: cache-opencv
-      id: cache-opencv
+    - name: cache-opencv-mobile
+      id: cache-opencv-mobile
       uses: actions/cache@v3
       with:
-        path: opencv
-        key: opencv-4.6.0
-    - name: opencv
-      if: steps.cache-opencv.outputs.cache-hit != 'true'
+        path: opencv-mobile
+        key: opencv-mobile-vs2019-v14
+    - name: opencv-mobile
+      if: steps.cache-opencv-mobile.outputs.cache-hit != 'true'
       run: |
-        Invoke-WebRequest -Uri https://github.com/opencv/opencv/releases/download/4.6.0/opencv-4.6.0-vc14_vc15.exe -OutFile opencv-4.6.0-vc14_vc15.exe
-        7z x ./opencv-4.6.0-vc14_vc15.exe
+        Invoke-WebRequest -Uri https://github.com/nihui/opencv-mobile/releases/download/v14/opencv-mobile-4.5.4-windows-vs2019.zip -OutFile opencv-mobile-4.5.4-windows-vs2019.zip
+        7z x ./opencv-mobile-4.5.4-windows-vs2019.zip
+        mv opencv-mobile-4.5.4-windows-vs2019 opencv-mobile
     - name: cache-ncnn
       id: cache-ncnn
       uses: actions/cache@v3
diff --git a/README.md b/README.md
index c7a4241..4f39d78 100644
--- a/README.md
+++ b/README.md
@@ -6,6 +6,13 @@
 
 ## 功能更新 Functional Update
 
+#### 2022/9/4
+
+- 修复了opencv-mobile写含bbox的图片时背景颜色错误的bug
+- 调整网络参数
+
+
+
 #### 2022/8/26
 
 - 初步适配须弥鱼类:更新了神经网络的结构与参数并进行初步测试,基本在须弥能够正常工作.
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 3848a50..76ad628 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -1,15 +1,18 @@
 cmake_minimum_required(VERSION 3.4.1)
 set(CMAKE_CXX_STANDARD 17)
 
-project(fisher VERSION 3.0.1)
+project(fisher VERSION 3.0.2)
 
 option(RELEASE "change directories" OFF)
 
 if(NOT RELEASE)
+    set(OpenCV_DIR "../../opencv/build") # collecting data needs cv::imshow
     option(TEST "collect rod data" OFF)
+else()
+    set(OpenCV_DIR "../../opencv-mobile/x64")
 endif()
 
-set(OpenCV_DIR "../../opencv/build")
+
 set(ncnn_INSTALL_DIR "../../ncnn/x64")
 set(ncnn_DIR "${ncnn_INSTALL_DIR}/lib/cmake/ncnn")
 set(ncnn_INCLUDE_DIR "${ncnn_INSTALL_DIR}/include")
diff --git a/src/fishing.cpp b/src/fishing.cpp
index 6fc44e5..39172b7 100644
--- a/src/fishing.cpp
+++ b/src/fishing.cpp
@@ -73,7 +73,14 @@ double colorDiff(const int refColor[], cv::Vec3b screenColor) {
 
 cv::Mat draw_bboxes(const cv::Mat &bgr, const std::vector<BoxInfo> &bboxes,
                     object_rect effect_roi) {
-  cv::Mat image = bgr.clone();
+    cv::Mat image;
+
+#ifdef RELEASE  // opencv-mobile reverse RGB and BGR here, but I don't know why
+  cv::cvtColor(bgr, image, cv::COLOR_BGR2RGB);
+#else
+  image = bgr.clone();
+#endif
+
   int src_w = image.cols;
   int src_h = image.rows;
   int dst_w = effect_roi.width;
diff --git a/src/fishing.h b/src/fishing.h
index 0fc5a1c..dfb493c 100644
--- a/src/fishing.h
+++ b/src/fishing.h
@@ -46,7 +46,7 @@ class Fisher {
   // unit: second
   const double MaxThrowWaiting = 3;
   const double MaxBiteWaiting[FISH_CLASS_NUM] = {
-      6, 8.5, 9.5, 10.5, 7, 10.5, 10, 10, 9, 8};  // index is fish label
+      8, 8.5, 9.5, 10.5, 8.5, 11.5, 11.5, 10.5, 9.5, 8.5};  // index is fish label
   const double MaxControlWaiting = 3;
 
   cv::Mat hookImg, pullImg, centralBarImg, leftEdgeImg, cursorImg, rightEdgeImg;
diff --git a/src/rodnet.cpp b/src/rodnet.cpp
index 4a132b0..0459139 100644
--- a/src/rodnet.cpp
+++ b/src/rodnet.cpp
@@ -3,29 +3,29 @@
 const double alpha =
     1734.34 / 2.5;  // tangent per pixel of nanodet input screen
 
-const double dz[FISH_CLASS_NUM] = {0.694767486335, 0.705337270051,
-                                   0.686332216768, 1.220813102738,
-                                   1.094953763608, 1.075469787444,
-                                   0.816349156903, 0.805647715939,
-                                   0.580136160296, 0.717789275813},
+const double dz[FISH_CLASS_NUM] = {0.593837091544, 0.638617811728,
+                                   0.686516414917, 1.101028742227,
+                                   1.045381575449, 1.092016629866,
+                                   0.910938338224, 1.537756036204,
+                                   0.795793167906, 1.009537510317},
              theta[3][1 + FISH_CLASS_NUM] =
-                 {{-0.063321387830, 0.271287522041, -0.289622288023,
-                   0.369747102726, -0.382151700443, -0.209881378546,
-                   0.216269445203, 0.047319592279, -0.336408813103,
-                   0.472667978506, 0.063048934754},
-                  {-0.409084259711, -0.435680067346, -0.021881190130,
-                   -0.029230749790, 0.452849262476, -0.152770124847,
-                   0.483840163995, 0.375548149249, 0.430277401509,
-                   -0.083056601200, 0.402188636327},
-                  {0.432541882301, 0.189117922865, 0.348296203478,
-                   -0.046222606584, 0.068145706124, 0.057344159790,
-                   -0.504384913277, -0.510813307110, -0.440694262582,
-                   -0.300708292908, -0.200549034557}},
-             B[3] = {0.864525353743, 1.243253422151,
-                     -2.503097924792};  // fitted parameters
-
-const double offset[FISH_CLASS_NUM] = {0.25, 0.05, 0.25, 0,   0.15,
-                                       0.2,  0.2,  0.1,  0.3, 0.15};
+                 {{0.031245754665, 0.408927414689, -0.236277908157,
+                   0.312233569326, -0.650706565353, 0.208001666710,
+                   0.134025076542, 0.022310548114, -0.528904220978,
+                   0.566728819567, 0.306536095251},
+                  {-0.452654476021, -0.591555546395, -0.286520879784,
+                   -0.204632958100, 0.271553442603, -0.046010366297,
+                   0.538585714359, 0.338996586166, 1.200797439567,
+                   0.057895275191, 0.356526749625},
+                  {0.594389625935, 0.055345825155, 0.294034879793,
+                   -0.265560742990, 0.093749871102, 0.143570197394,
+                   -0.797929347925, -0.683466901431, -0.694927314353,
+                   -0.548503360543, -0.782646406893}},
+             B[3] = {0.789249430622, 1.812164387601,
+                     -2.925873549802};  // fitted parameters
+
+const double offset[FISH_CLASS_NUM] = {0.3, 0.1, 0.25, 0,   0.2,
+                                       0.2, 0.2, -0.1, 0.3, 0.25};
 
 // dst = {a,b,v}
 void f(double* dst, double* x, double* y) {