diff --git a/app/build.gradle b/app/build.gradle
index 0f17218cd..ab769cfb9 100644
--- a/app/build.gradle
+++ b/app/build.gradle
@@ -64,7 +64,7 @@ android {
             applicationIdSuffix ".debug"
         }
         release {
-//            debuggable true
+            debuggable false
             shrinkResources false
             minifyEnabled false
             proguardFiles getDefaultProguardFile('proguard-android.txt'), 'proguard-rules.pro'
diff --git a/app/proguard-rules.pro b/app/proguard-rules.pro
index 6ccb2012b..0e3606353 100644
--- a/app/proguard-rules.pro
+++ b/app/proguard-rules.pro
@@ -28,11 +28,14 @@
 -dontwarn java.lang.invoke.*
 -dontwarn **$$Lambda$*
 
+-keepnames class com.stardust.autojs.onnx.YoloV8Predictor
+
 -keep class org.mozilla.javascript.** { *; }
 -keep class com.jecelyin.editor.** { *; }
 -keep class com.stardust.automator.** { *; }
 -keep class com.stardust.autojs.** { *; }
 -keep class org.greenrobot.eventbus.** { *; }
+-keep class org.autojs.autojs.** {*;}
 -keep class * extends c
 -keepattributes *Annotation*
 # Event bus
@@ -138,4 +141,16 @@
 # Bugly
 
 -dontwarn com.tencent.bugly.**
--keep public class com.tencent.bugly.**{*;}
\ No newline at end of file
+-keep public class com.tencent.bugly.**{*;}
+
+# OnnxRuntime
+-dontwarn ai.onnxruntime.**
+-keep public class ai.onnxruntime.**{*;}
+
+# okHttp3
+-dontwarn okhttp3.**
+-keep public class okhttp3.**{*;}
+
+# paddleocr
+-dontwarn com.baidu.paddle.lite.ocr.**
+-keep public class com.baidu.paddle.lite.ocr.**{*;}
\ No newline at end of file
diff --git a/app/src/main/AndroidManifest.xml b/app/src/main/AndroidManifest.xml
index 662d32c5b..7b707473c 100644
--- a/app/src/main/AndroidManifest.xml
+++ b/app/src/main/AndroidManifest.xml
@@ -45,7 +45,9 @@
         android:requestLegacyExternalStorage="true"
         tools:replace="android:label, android:icon, android:allowBackup"
         tools:targetApi="m">
-
+        <!-- 开启内存分析 -->
+        <profileable android:shell="true" android:enabled="true"
+            tools:targetApi="q" />
         <meta-data
             android:name="android.max_aspect"
             android:value="2.1"/>
diff --git a/autojs-aar/paddleocr/src/main/cpp/ocr_db_post_process.cpp b/autojs-aar/paddleocr/src/main/cpp/ocr_db_post_process.cpp
index 9816ea4ac..a2cf342f9 100644
--- a/autojs-aar/paddleocr/src/main/cpp/ocr_db_post_process.cpp
+++ b/autojs-aar/paddleocr/src/main/cpp/ocr_db_post_process.cpp
@@ -64,6 +64,13 @@ static cv::RotatedRect unclip(float **box) {
 
   return res;
 }
+// 用于动态释放内存 避免内存泄露
+static void free_array(float **array, int rows) {
+  for (int i = 0; i < rows; ++i) {
+    delete[] array[i];
+  }
+  delete[] array;
+}
 
 static float **Mat2Vec(cv::Mat mat) {
   auto **array = new float *[mat.rows];
@@ -267,6 +274,7 @@ boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap) {
     // end get_mini_box
 
     if (ssid < min_size) {
+      free_array(array, 4);
       continue;
     }
 
@@ -274,6 +282,7 @@ boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap) {
     score = box_score_fast(array, pred);
     // end box_score_fast
     if (score < box_thresh) {
+      free_array(array, 4);
       continue;
     }
 
@@ -284,8 +293,11 @@ boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap) {
     cv::RotatedRect clipbox = points;
     auto cliparray = get_mini_boxes(clipbox, ssid);
 
-    if (ssid < min_size + 2)
+    if (ssid < min_size + 2) {
+      free_array(array, 4);
+      free_array(cliparray, 4);
       continue;
+    }
 
     int dest_width = pred.cols;
     int dest_height = pred.rows;
@@ -301,6 +313,8 @@ boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap) {
       intcliparray.emplace_back(std::move(a));
     }
     boxes.emplace_back(std::move(intcliparray));
+    free_array(array, 4);
+    free_array(cliparray, 4);
 
   } // end for
   return boxes;
diff --git a/autojs/build.gradle b/autojs/build.gradle
index 7b516be01..b2e51b0bf 100644
--- a/autojs/build.gradle
+++ b/autojs/build.gradle
@@ -15,7 +15,7 @@ android {
     }
     buildTypes {
         release {
-            minifyEnabled false
+            minifyEnabled true
             proguardFiles getDefaultProguardFile('proguard-android.txt'), 'proguard-rules.pro'
         }
     }
diff --git a/autojs/proguard-rules.pro b/autojs/proguard-rules.pro
index 27b6816cb..e588db103 100644
--- a/autojs/proguard-rules.pro
+++ b/autojs/proguard-rules.pro
@@ -23,3 +23,5 @@
 # If you keep the line number information, uncomment this to
 # hide the original source file name.
 #-renamesourcefileattribute SourceFile
+#-keep public class com.stardust.autojs.onnx.YoloV8Predictor
+#-keepnames class com.stardust.autojs.onnx.YoloV8Predictor
diff --git a/autojs/src/main/assets/modules/__$ocr__.js b/autojs/src/main/assets/modules/__$ocr__.js
index 055deaa08..de0890308 100644
--- a/autojs/src/main/assets/modules/__$ocr__.js
+++ b/autojs/src/main/assets/modules/__$ocr__.js
@@ -31,11 +31,13 @@ module.exports = function(runtime, global) {
     let region = options.region
     if (region) {
       let r = buildRegion(region, img)
-      let o = img
       img = images.clip(img, r.x, r.y, r.width, r.height)
-      o.recycle()
     }
     let text = javaOcr.recognizeText(img, options.cpuThreadNum || 4, options.useSlim || true)
+    if (region) {
+      // 进行过区域截取，需要回收截取的图片 原始图片由外部管理
+      img.recycle()
+    }
     if (text) {
       return JSON.parse(JSON.stringify(text))
     }
@@ -47,14 +49,16 @@ module.exports = function(runtime, global) {
     let region = options.region
     if (region) {
       let r = buildRegion(region, img)
-      let o = img
       img = images.clip(img, r.x, r.y, r.width, r.height)
-      o.recycle()
     }
     let resultList = runtime.bridges.bridges.toArray(javaOcr.detect(img, options.cpuThreadNum || 4, options.useSlim || true))
     if (region && region.length > 1 && resultList && resultList.length > 0) {
       resultList.forEach(r => r.bounds.offset(region[0], region[1]))
     }
+    if (region) {
+      // 进行过区域截取，需要回收截取的图片 原始图片由外部管理
+      img.recycle()
+    }
     return resultList
   }
 
diff --git a/autojs/src/main/assets/modules/__$yolo__.js b/autojs/src/main/assets/modules/__$yolo__.js
index ac14e37dd..9e6aaaf80 100644
--- a/autojs/src/main/assets/modules/__$yolo__.js
+++ b/autojs/src/main/assets/modules/__$yolo__.js
@@ -1,5 +1,6 @@
 module.exports = function (runtime, global) {
   let yoloCreator = runtime.yolo
+  let instanceList = []
   let $yolo = function () {
     this.enabled = false
   }
@@ -29,6 +30,7 @@ module.exports = function (runtime, global) {
       }
       this.yoloInstance = yoloCreator.createNcnn(options.paramPath, options.binPath, convertToList(options.labels), options.imageSize, !!options.useGpu);
       this.enabled = this.yoloInstance.isInit();
+      instanceList.push(this.yoloInstance)
     } else {
       this.type = 'onnx';
 
@@ -39,6 +41,7 @@ module.exports = function (runtime, global) {
       }
       this.yoloInstance = yoloCreator.createOnnx(options.modelPath, convertToList(options.labels), options.imageSize);
       this.enabled = this.yoloInstance != null;
+      instanceList.push(this.yoloInstance)
     }
 
     if (this.enabled) {
@@ -60,9 +63,7 @@ module.exports = function (runtime, global) {
     }
     if (region) {
       let r = buildRegion(region, img)
-      let o = img
       img = images.clip(img, r.x, r.y, r.width, r.height)
-      o.recycle()
     }
     let resultList = util.java.toJsArray(this.yoloInstance.predictYolo(img.mat))
     if (region) {
@@ -94,6 +95,12 @@ module.exports = function (runtime, global) {
 
   return $yolo;
 
+  events.on('exit', function () {
+    if (instanceList.length > 0) {
+        instanceList.forEach(instance => instance.release())
+    }
+  })
+
   function wrapForward (yoloInstance) {
     return {
       forward: function (img, filterOption, region) {
diff --git a/autojs/src/main/assets/modules/__engines__.js b/autojs/src/main/assets/modules/__engines__.js
index a8104e42c..1346c47d5 100644
--- a/autojs/src/main/assets/modules/__engines__.js
+++ b/autojs/src/main/assets/modules/__engines__.js
@@ -1,47 +1,45 @@
 
-module.exports = function(__runtime__, scope){
+module.exports = function (__runtime__, scope) {
     var rtEngines = __runtime__.engines;
-    var execArgv = null;
-
     var engines = {};
 
-    engines.execScript = function(name, script, config){
+    engines.execScript = function (name, script, config) {
         return rtEngines.execScript(name, script, fillConfig(config));
     }
 
-    engines.execScriptFile = function(path, config){
+    engines.execScriptFile = function (path, config) {
         return rtEngines.execScriptFile(path, fillConfig(config));
     }
 
-    engines.execAutoFile = function(path, config){
+    engines.execAutoFile = function (path, config) {
         return rtEngines.execAutoFile(path, fillConfig(config));
     }
 
-    engines.myEngine = function(){
+    engines.myEngine = function () {
         return rtEngines.myEngine();
     }
 
-    engines.all = function(){
+    engines.all = function () {
         return rtEngines.all();
     }
 
     engines.stopAll = rtEngines.stopAll.bind(rtEngines);
     engines.stopAllAndToast = rtEngines.stopAllAndToast.bind(rtEngines);
 
-    function fillConfig(c){
+    function fillConfig (c) {
         var config = new com.stardust.autojs.execution.ExecutionConfig();
         c = c || {};
         c.path = c.path || files.cwd();
-        if(c.path){
-           config.workingDirectory = c.path;
+        if (c.path) {
+            config.workingDirectory = c.path;
         }
         config.delay = c.delay || 0;
         config.interval = c.interval || 0;
-        config.loopTimes = (c.loopTimes === undefined)? 1 : c.loopTimes;
-        if(c.arguments){
+        config.loopTimes = (c.loopTimes === undefined) ? 1 : c.loopTimes;
+        if (c.arguments) {
             var arguments = c.arguments;
-            for(var key in arguments){
-                if(arguments.hasOwnProperty(key)){
+            for (var key in arguments) {
+                if (arguments.hasOwnProperty(key)) {
                     config.setArgument(key, arguments[key]);
                 }
             }
@@ -49,14 +47,15 @@ module.exports = function(__runtime__, scope){
         return config;
     }
 
-    var engine = engines.myEngine();
-    var execArgv = {};
-    var iterator = engine.getTag("execution.config").arguments.entrySet().iterator();
-    while(iterator.hasNext()){
-        var entry = iterator.next();
-        execArgv[entry.getKey()] = entry.getValue();
-    }
-    engine.execArgv = execArgv;
+    ((engine) => {
+        let execArgv = {}
+        let iterator = engine.getTag("execution.config").arguments.entrySet().iterator();
+        while (iterator.hasNext()) {
+            let entry = iterator.next();
+            execArgv[entry.getKey()] = entry.getValue();
+        }
+        engine.execArgv = execArgv
+    })(engines.myEngine())
 
     return engines;
 }
\ No newline at end of file
diff --git a/autojs/src/main/java/com/stardust/autojs/core/image/ImageWrapper.java b/autojs/src/main/java/com/stardust/autojs/core/image/ImageWrapper.java
index 846850a76..8b5f9da3a 100644
--- a/autojs/src/main/java/com/stardust/autojs/core/image/ImageWrapper.java
+++ b/autojs/src/main/java/com/stardust/autojs/core/image/ImageWrapper.java
@@ -108,49 +108,49 @@ public static Bitmap toBitmap(Image image) {
     }
 
     public static ImageWrapper ofImageByMat(Image image, int cvType) {
-
         long start = System.currentTimeMillis();
 
         // 获取Image的平面
         Image.Plane[] planes = image.getPlanes();
-        Log.d("ImageWrapper", "ofImageByMat: planes.length: " + planes.length);
+//        Log.d("ImageWrapper", "ofImageByMat: planes.length: " + planes.length);
         // 获取Image的宽高
         int width = image.getWidth();
         int height = image.getHeight();
+//        Log.d("ImageWrapper", "ofImageByMat: width:" + width + " height:" + height);
 
-        Log.d("ImageWrapper", "ofImageByMat: width:" + width + " height:" + height);
         Image.Plane plane = planes[0];
         // 获取平面的数据缓冲区
-        ByteBuffer buffer = planes[0].getBuffer();
-        // 创建一个byte数组来存储缓冲区的数据
-        byte[] bytes = new byte[buffer.remaining()];
-        buffer.position(0);
+        ByteBuffer buffer = plane.getBuffer();
         int pixelStride = plane.getPixelStride();
-        int rowPadding = plane.getRowStride() - pixelStride * image.getWidth();
-        // 将数据从缓冲区拷贝到byte数组
-        buffer.get(bytes);
+        int rowStride = plane.getRowStride();
+        int rowPadding = rowStride - pixelStride * width;
 
-        // 创建一个Mat对象
+        long s2 = System.currentTimeMillis();
+        // 尽量避免使用临时数组
         Mat mat = new Mat(height, width + rowPadding / pixelStride, CvType.CV_8UC4);
-        // 将byte数组拷贝到Mat对象
-        mat.put(0, 0, bytes);
+        byte[] rowData = new byte[rowStride];
+        for (int i = 0; i < height; i++) {
+            buffer.get(rowData, 0, rowStride);
+            mat.put(i, 0, rowData);
+        }
+//        Log.d("ImageWrapper", "ofImageByMat: create mat by bytes cost: " + (System.currentTimeMillis() - s2) + "ms");
         if (width != mat.width()) {
-            Log.d("ImageWrapper", "ofImageByMat: mat width is not valid: " + mat.width() + " => " + width);
+//            Log.d("ImageWrapper", "ofImageByMat: mat width is not valid: " + mat.width() + " => " + width);
             // 定义裁切区域
             Rect rect = new Rect(0, 0, width, height);
-            // 裁切图像
             Mat croppedImage = new Mat(mat, rect);
             mat.release();
             mat = croppedImage;
         }
+
         if (cvType != CvType.CV_8UC4) {
             long convertStart = System.currentTimeMillis();
             Imgproc.cvtColor(mat, mat, Imgproc.COLOR_RGBA2RGB);
-            Log.d("ImageWrapper", "ofImageByMat: convert channel: " + (System.currentTimeMillis() - convertStart) + "ms");
+//            Log.d("ImageWrapper", "ofImageByMat: convert channel: " + (System.currentTimeMillis() - convertStart) + "ms");
         }
-        Log.d("ImageWrapper", "ofImageByMat: create by mat cost: " + (System.currentTimeMillis() - start) + "ms");
-        return new ImageWrapper(mat);
 
+//        Log.d("ImageWrapper", "ofImageByMat: create by mat cost: " + (System.currentTimeMillis() - start) + "ms");
+        return new ImageWrapper(mat);
     }
 
     public int getWidth() {
diff --git a/autojs/src/main/java/com/stardust/autojs/execution/RunnableScriptExecution.java b/autojs/src/main/java/com/stardust/autojs/execution/RunnableScriptExecution.java
index 19d9d2bf3..c8775e1b0 100644
--- a/autojs/src/main/java/com/stardust/autojs/execution/RunnableScriptExecution.java
+++ b/autojs/src/main/java/com/stardust/autojs/execution/RunnableScriptExecution.java
@@ -2,7 +2,6 @@
 
 import android.util.Log;
 
-import com.stardust.autojs.AutoJs;
 import com.stardust.autojs.ScriptEngineService;
 import com.stardust.autojs.engine.ScriptEngine;
 import com.stardust.autojs.engine.ScriptEngineManager;
diff --git a/autojs/src/main/java/com/stardust/autojs/onnx/YoloV8Predictor.java b/autojs/src/main/java/com/stardust/autojs/onnx/YoloV8Predictor.java
index f063744f4..538003aab 100644
--- a/autojs/src/main/java/com/stardust/autojs/onnx/YoloV8Predictor.java
+++ b/autojs/src/main/java/com/stardust/autojs/onnx/YoloV8Predictor.java
@@ -9,6 +9,7 @@
 import com.stardust.autojs.onnx.util.Letterbox;
 import com.stardust.autojs.runtime.api.YoloPredictor;
 
+import org.opencv.core.CvType;
 import org.opencv.core.Mat;
 import org.opencv.core.Size;
 import org.opencv.imgcodecs.Imgcodecs;
@@ -17,11 +18,14 @@
 import java.nio.FloatBuffer;
 import java.util.ArrayList;
 import java.util.Arrays;
+import java.util.Collections;
 import java.util.Comparator;
 import java.util.EnumSet;
 import java.util.HashMap;
 import java.util.List;
 import java.util.Map;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
 import java.util.stream.Collectors;
 
 import ai.onnxruntime.OnnxTensor;
@@ -39,6 +43,8 @@
 @RequiresApi(api = Build.VERSION_CODES.N)
 public class YoloV8Predictor extends YoloPredictor {
     private static final String TAG = "YoloV8Predictor";
+    private static final Pattern IMG_SIZE_PATTERN = Pattern.compile("\\[(\\d+), \\d+]");
+    private static final Pattern LABEL_PATTERN = Pattern.compile("'([^']*)'");
 
     private final String modelPath;
 
@@ -95,8 +101,53 @@ private void prepareSession() throws OrtException {
                 throw new RuntimeException(e);
             }
         });
+        // 如果入参labels无效或未定义，使用模型内置labels
+        if (labels == null || labels.size() == 0) {
+            labels = initLabels(session);
+        }
+        initShapeSize(session);
+    }
+
+    private List<String> initLabels(OrtSession session) throws OrtException {
+        String meteStr = session.getMetadata().getCustomMetadata().get("names");
+        if (meteStr == null) {
+            Log.d(TAG, "initLabels: 读取names失败 无法自动修正labels");
+            return Collections.emptyList();
+        }
+        String[] labels = new String[meteStr.split(",").length];
+
+        Matcher matcher = LABEL_PATTERN.matcher(meteStr);
+
+        int h = 0;
+        while (matcher.find()) {
+            labels[h] = matcher.group(1);
+            h++;
+        }
+        return Arrays.asList(labels);
     }
 
+    private void initShapeSize(OrtSession session) throws OrtException {
+        String meteStr = session.getMetadata().getCustomMetadata().get("imgsz");
+        Log.d(TAG, "initShapeSize: " + meteStr);
+        if (meteStr == null) {
+            Log.d(TAG, "initShapeSize: 读取imgsz失败 无法自动修正输入大小");
+            return;
+        }
+        Matcher matcher = IMG_SIZE_PATTERN.matcher(meteStr);
+        if (matcher.find()) {
+            String shapeSize = matcher.group(1);
+            if (shapeSize == null) {
+                Log.d(TAG, "initShapeSize: 读取imgsz格式异常 无法自动修正输入大小");
+                return;
+            }
+            this.shapeSize = new Size(Double.parseDouble(shapeSize), Double.parseDouble(shapeSize));
+            Log.d(TAG, "set shape size: " + shapeSize);
+        } else {
+            Log.d(TAG, "initShapeSize: 读取imgsz格式异常 无法自动修正输入大小");
+        }
+    }
+
+
     private void addNNApiProvider(OrtSession.SessionOptions sessionOptions) {
         if (!tryNpu) {
             return;
@@ -140,19 +191,26 @@ private HashMap<String, OnnxTensor> preprocessImage(Mat img) throws OrtException
         int rows = letterbox.getHeight();
         int cols = letterbox.getWidth();
         int channels = image.channels();
+        // 转换Mat对象的数据类型为CV_64F，即64位浮点型
+        Mat convertedImage = new Mat();
+        image.convertTo(convertedImage, CvType.CV_64F);
+
+        // 获取整个像素数据
+        double[] pixelData = new double[rows * cols * channels];
+        convertedImage.get(0, 0, pixelData);
 
-        // 将Mat对象的像素值赋值给Float[]对象
         float[] pixels = new float[channels * rows * cols];
         for (int i = 0; i < rows; i++) {
             for (int j = 0; j < cols; j++) {
-                double[] pixel = image.get(j, i);
                 for (int k = 0; k < channels; k++) {
                     // 这样设置相当于同时做了image.transpose((2, 0, 1))操作
-                    pixels[rows * cols * k + j * cols + i] = (float) pixel[k] / 255.0f;
+                    // 重新组织内存访问模式，提高缓存效率
+                    pixels[k * rows * cols + i * cols + j] = (float) (pixelData[(i * cols + j) * channels + k] / 255.0);
                 }
             }
         }
         image.release();
+        convertedImage.release();
         // 创建OnnxTensor对象
         long[] shape = {1L, (long) channels, (long) rows, (long) cols};
         OnnxTensor tensor = OnnxTensor.createTensor(environment, FloatBuffer.wrap(pixels), shape);
@@ -168,9 +226,8 @@ private List<Detection> postProcessOutput(OrtSession.Result output) throws OrtEx
         Map<Integer, List<float[]>> class2Bbox = new HashMap<>();
 
         for (float[] bbox : outputData) {
-            float[] conditionalProbabilities = Arrays.copyOfRange(bbox, 4, outputData.length);
-            int label = argmax(conditionalProbabilities);
-            float conf = conditionalProbabilities[label];
+            int label = argmax(bbox, 4); // 直接在原数组上进行操作
+            float conf = bbox[label + 4];
             if (conf < confThreshold) {
                 continue;
             }
@@ -185,8 +242,7 @@ private List<Detection> postProcessOutput(OrtSession.Result output) throws OrtEx
                 continue;
             }
 
-            class2Bbox.putIfAbsent(label, new ArrayList<>());
-            class2Bbox.get(label).add(bbox);
+            class2Bbox.computeIfAbsent(label, k -> new ArrayList<>()).add(bbox);
         }
 
         List<Detection> detections = new ArrayList<>();
@@ -214,12 +270,18 @@ public List<DetectResult> predictYolo(String imagePath) throws OrtException {
     public List<DetectResult> predictYolo(Mat image) throws OrtException {
         prepareSession();
         long start_time = System.currentTimeMillis();
+        Map<String, OnnxTensor> inputMap = preprocessImage(image);
         // 运行推理
-        OrtSession.Result output = session.run(preprocessImage(image));
-        List<Detection> detections = postProcessOutput(output);
-        Log.d("YoloV8Predictor", String.format("onnx predict cost: %d ms", (System.currentTimeMillis() - start_time)));
-        return detections.stream().map(detection -> new DetectResult(detection, letterbox))
-                .collect(Collectors.toList());
+        try (OrtSession.Result output = session.run(inputMap)) {
+            Log.d(TAG, "predictYolo: onnx run cost " + (System.currentTimeMillis() - start_time) + "ms");
+            List<Detection> detections = postProcessOutput(output);
+            Log.d("YoloV8Predictor", String.format("onnx predict cost: %d ms", (System.currentTimeMillis() - start_time)));
+            return detections.stream().map(detection -> new DetectResult(detection, letterbox))
+                    .collect(Collectors.toList());
+        } finally {
+            // 释放资源
+            inputMap.values().forEach(OnnxTensor::close);
+        }
     }
 
     public static void xywh2xyxy(float[] bbox) {
@@ -245,7 +307,7 @@ public static float[][] transposeMatrix(float[][] m) {
     }
 
     public static List<float[]> nonMaxSuppression(List<float[]> bboxes, float iouThreshold) {
-
+        long start = System.currentTimeMillis();
         List<float[]> bestBboxes = new ArrayList<>();
 
         bboxes.sort(Comparator.comparing(a -> a[4]));
@@ -253,9 +315,9 @@ public static List<float[]> nonMaxSuppression(List<float[]> bboxes, float iouThr
         while (!bboxes.isEmpty()) {
             float[] bestBbox = bboxes.remove(bboxes.size() - 1);
             bestBboxes.add(bestBbox);
-            bboxes = bboxes.stream().filter(a -> computeIOU(a, bestBbox) < iouThreshold).collect(Collectors.toList());
+            bboxes.removeIf(bbox -> computeIOU(bbox, bestBbox) >= iouThreshold);
         }
-
+        Log.d(TAG, "nonMaxSuppression: cost " + (System.currentTimeMillis() - start) + "ms");
         return bestBboxes;
     }
 
@@ -269,22 +331,44 @@ public static float computeIOU(float[] box1, float[] box2) {
         float right = Math.min(box1[2], box2[2]);
         float bottom = Math.min(box1[3], box2[3]);
 
-        float interArea = Math.max(right - left, 0) * Math.max(bottom - top, 0);
+        // 计算交集区域的宽度和高度
+        float width = Math.max(right - left, 0);
+        float height = Math.max(bottom - top, 0);
+
+        // 计算交集面积和并集面积
+        float interArea = width * height;
         float unionArea = area1 + area2 - interArea;
-        return Math.max(interArea / unionArea, 1e-8f);
 
+        // 计算交并比
+        return Math.max(interArea / unionArea, 1e-8f);
     }
 
+
     //返回最大值的索引
-    public static int argmax(float[] a) {
+    // 优化后的 argmax 函数
+    public static int argmax(float[] a, int start) {
         float re = -Float.MAX_VALUE;
         int arg = -1;
-        for (int i = 0; i < a.length; i++) {
+        for (int i = start; i < a.length; i++) {
             if (a[i] >= re) {
                 re = a[i];
-                arg = i;
+                arg = i - start;
             }
         }
         return arg;
     }
+
+    @Override
+    public void release() {
+        if (session != null) {
+            try {
+                session.close();
+                session = null;
+            } catch (OrtException e) {
+                Log.e(TAG, "close session failed" + e);
+            }
+            environment.close();
+            environment = null;
+        }
+    }
 }
diff --git a/autojs/src/main/java/com/stardust/autojs/runtime/api/Images.java b/autojs/src/main/java/com/stardust/autojs/runtime/api/Images.java
index 0f1945b65..68fa24c3e 100644
--- a/autojs/src/main/java/com/stardust/autojs/runtime/api/Images.java
+++ b/autojs/src/main/java/com/stardust/autojs/runtime/api/Images.java
@@ -118,6 +118,7 @@ public synchronized ImageWrapper captureScreen() {
         }
         long start = System.currentTimeMillis();
         mPreCaptureImage = ImageWrapper.ofImage(capture);
+        capture.close();
         Log.d(TAG, "captureScreen: convert image cost: " + (System.currentTimeMillis() - start) + "ms");
         return mPreCaptureImage;
     }
diff --git a/autojs/src/main/java/com/stardust/autojs/runtime/api/Yolo.java b/autojs/src/main/java/com/stardust/autojs/runtime/api/Yolo.java
index ad8ffd8f8..ee70d19e2 100644
--- a/autojs/src/main/java/com/stardust/autojs/runtime/api/Yolo.java
+++ b/autojs/src/main/java/com/stardust/autojs/runtime/api/Yolo.java
@@ -103,7 +103,8 @@ public List<DetectResult> captureAndPredict(ScriptRuntime runtime, Rect rect) {
             Images images = (Images)runtime.getImages();
             Image image = images.captureScreenRaw();
             if (image != null) {
-                ImageWrapper imageWrapper = ImageWrapper.ofImageByMat(image, CvType.CV_8UC3);
+                ImageWrapper imageWrapper = ImageWrapper.ofImageByMat(image, CvType.CV_8UC4);
+                image.close();
                 Mat mat = imageWrapper.getMat();
                 if (rect != null) {
                     // 裁切图像
diff --git a/autojs/src/main/java/com/stardust/autojs/runtime/api/YoloPredictor.java b/autojs/src/main/java/com/stardust/autojs/runtime/api/YoloPredictor.java
index bf514c88d..c4b62723b 100644
--- a/autojs/src/main/java/com/stardust/autojs/runtime/api/YoloPredictor.java
+++ b/autojs/src/main/java/com/stardust/autojs/runtime/api/YoloPredictor.java
@@ -57,4 +57,10 @@ public boolean isInit() {
     public void release() {
 
     }
+
+    @Override
+    protected void finalize() throws Throwable {
+        super.finalize();
+        release();
+    }
 }