优化自定义分类器：实现未知类别检测的二阶段策略

// ... (图像预处理代码) ...

FruitDisease.Outputs outputs = model.process(inputFeature);
TensorBuffer outputFeature0 = outputs.getOutputFeature0AsTensorBuffer();

float[] confidences = outputFeature0.getFloatArray();
int maxPos = 0;
float maxConfidence = 0;
for (int i = 0; i < confidences.length; i++) {
    if (confidences[i] > maxConfidence) {
        maxConfidence = confidences[i];
        maxPos = i;
    }
}

String[] classes = {"Watermelon Healthy", "Watermelon Blossom End Rot", "Watermelon Anthracnose", /* ... 其他类别 ... */};

result.setText(classes[maxPos]); // 总是会显示一个类别

private void classifyImage(Bitmap image) {
    try {
        // 1. 图像预处理 (与原代码相同)
        // ... (省略预处理细节) ...
        TensorBuffer inputFeature = TensorBuffer.createFixedSize(new int[]{1, 224, 224, 3}, DataType.FLOAT32);
        // ... (加载图片到 inputFeature) ...

        // 阶段一：二元分类 - 检测是否存在目标 (例如，是否存在水果)
        // 假设您有一个名为 FruitPresenceModel 的二元分类模型
        FruitPresenceModel presenceModel = FruitPresenceModel.newInstance(getApplicationContext());
        FruitPresenceModel.Outputs presenceOutputs = presenceModel.process(inputFeature);
        TensorBuffer presenceOutputBuffer = presenceOutputs.getOutputFeature0AsTensorBuffer(); // 假设输出是 [1, 2] 或 [1, 1]
        float[] presenceConfidences = presenceOutputBuffer.getFloatArray();

        // 假设 presenceConfidences[0] 是“非水果”的置信度，presenceConfidences[1] 是“水果”的置信度
        // 或者如果模型只输出一个值，比如“水果”的概率
        float fruitProbability = presenceConfidences.length > 1 ? presenceConfidences[1] : presenceConfidences[0]; // 根据模型实际输出调整

        float presenceThreshold = 0.7f; // 设置一个阈值，判断是否为水果

        if (fruitProbability > presenceThreshold) {
            // 阶段二：多类别分类 - 识别具体是哪种水果
            FruitDisease multiClassModel = FruitDisease.newInstance(getApplicationContext());
            FruitDisease.Outputs multiClassOutputs = multiClassModel.process(inputFeature);
            TensorBuffer multiClassOutputBuffer = multiClassOutputs.getOutputFeature0AsTensorBuffer();

            float[] confidences = multiClassOutputBuffer.getFloatArray();
            int maxPos = 0;
            float maxConfidence = 0;
            for (int i = 0; i < confidences.length; i++) {
                if (confidences[i] > maxConfidence) {
                    maxConfidence = confidences[i];
                    maxPos = i;
                }
            }

            // 再次检查多类别分类的置信度，确保不是一个非常低的预测
            float multiClassConfidenceThreshold = 0.6f; // 可以根据实际情况调整

            String[] classes = {"Watermelon Healthy", "Watermelon Blossom End Rot", /* ... 其他水果类别 ... */};

            if (maxConfidence > multiClassConfidenceThreshold) {
                result.setText(classes[maxPos]);
                // 显示所有类别的置信度
                StringBuilder s = new StringBuilder();
                for (int i = 0; i < classes.length; i++) {
                    s.append(String.format("%s: %.1f%%\n", classes[i], confidences[i] * 100));
                }
                confidence.setText(s.toString());
                confidence.setVisibility(View.VISIBLE);
            } else {
                // 虽然第一阶段认为是水果，但第二阶段的置信度太低，可能是模糊或难以识别的图片
                result.setText("未检测到明确的水果类别。");
                confidence.setVisibility(View.GONE);
            }

        } else {
            // 第一阶段判断为非水果
            result.setText("未检测到水果。");
            confidence.setVisibility(View.GONE);
        }

        // 释放模型资源
        presenceModel.close();
        // 如果在条件块内创建，则在条件块内关闭，或者确保在finally块中关闭所有模型
        // multiClassModel.close(); // 需要根据实际模型生命周期管理
    } catch (Exception e) {
        // 处理异常
        result.setText("分类失败：" + e.getMessage());
        confidence.setVisibility(View.GONE);
    }
}