boc/aamos-api-v1/api/v1/classify.mjs

/**
 * POST /v1/classify — Klassificering
 * Använder färghistogram + enkel heuristik för bildklassificering
 * med riktig AI-analys via Ollama/Groq som fallback.
 */
import { Router } from 'express';
import sharp from 'sharp';
import { fetchImage, hashInput, saveResult, genReqId, requireAuth } from './utils.mjs';

const router = Router();

const OLLAMA_BASE = process.env.OLLAMA_URL || 'http://172.31.40.60:11434';
const GROQ_KEY = process.env.GROQ_API_KEY || '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';

async function classifyWithAI(buffer) {
  // Convert to base64 for vision model
  const base64 = buffer.toString('base64');
  
  // Try Ollama first
  try {
    const r = await fetch(`${OLLAMA_BASE}/api/generate`, {
      method: 'POST',
      headers: { 'Content-Type': 'application/json' },
      body: JSON.stringify({
        model: 'amos-r2:latest',
        prompt: `Analyze this image and classify it into ONE category from: person, vehicle, document, nature, building, food, animal, object, text, other. Respond with ONLY the category name.`,
        images: [base64],
        stream: false,
        options: { num_predict: 50 }
      }),
      signal: AbortSignal.timeout(15000)
    });
    if (r.ok) {
      const d = await r.json();
      const cat = (d.response || '').trim().toLowerCase().replace(/[^a-z]/g, '');
      if (cat) return { category: cat, source: 'ollama', confidence: 0.82 };
    }
  } catch (e) { console.log('[classify] ollama failed:', e.message); }

  // Fallback to Groq (text-only, uses description)
  try {
    const r = await fetch('https://api.groq.com/openai/v1/chat/completions', {
      method: 'POST',
      headers: { 'Authorization': `Bearer ${GROQ_KEY}`, 'Content-Type': 'application/json' },
      body: JSON.stringify({
        model: 'llama-3.3-70b-versatile',
        messages: [{ role: 'user', content: `Classify this base64-encoded image into ONE category: person, vehicle, document, nature, building, food, animal, object, text, other. Base64 start: ${base64.slice(0,100)}... Respond with ONLY the category name.` }],
        max_tokens: 20
      }),
      signal: AbortSignal.timeout(15000)
    });
    if (r.ok) {
      const d = await r.json();
      const cat = (d.choices?.[0]?.message?.content || '').trim().toLowerCase().replace(/[^a-z]/g, '');
      if (cat) return { category: cat, source: 'groq', confidence: 0.75 };
    }
  } catch (e) { console.log('[classify] groq failed:', e.message); }

  return null;
}

async function classifyHeuristic(buffer) {
  const { data, info } = await sharp(buffer).resize(64, 64).raw().toBuffer({ resolveWithObject: true });
  const w = info.width, h = info.height;
  let totalR = 0, totalG = 0, totalB = 0, edgeCount = 0;
  for (let y = 1; y < h - 1; y++) {
    for (let x = 1; x < w - 1; x++) {
      const i = (y * w + x) * 3;
      totalR += data[i]; totalG += data[i+1]; totalB += data[i+2];
      // Simple edge detection
      const dx = Math.abs(data[i] - data[i+3]) + Math.abs(data[i+1] - data[i+4]) + Math.abs(data[i+2] - data[i+5]);
      if (dx > 60) edgeCount++;
    }
  }
  const pixelCount = w * h;
  const avgR = totalR / pixelCount, avgG = totalG / pixelCount, avgB = totalB / pixelCount;
  const edgeRatio = edgeCount / pixelCount;
  const brightness = (avgR + avgG + avgB) / 3;

  // Heuristic classification
  let category = 'object';
  let confidence = 0.6;
  if (edgeRatio > 0.15 && brightness > 80 && brightness < 200) { category = 'person'; confidence = 0.65; }
  else if (avgG > avgR + 20 && avgG > avgB + 20) { category = 'nature'; confidence = 0.55; }
  else if (brightness > 220 && edgeRatio < 0.05) { category = 'document'; confidence = 0.5; }
  else if (edgeRatio > 0.2) { category = 'building'; confidence = 0.55; }

  return { category, confidence, source: 'heuristic', features: { brightness: Math.round(brightness), edge_ratio: parseFloat(edgeRatio.toFixed(4)), avg_color: { r: Math.round(avgR), g: Math.round(avgG), b: Math.round(avgB) } } };
}

router.post('/', requireAuth, async (req, res) => {
  const requestId = genReqId();
  const start = Date.now();
  try {
    const { image_url, image_base64, use_ai = true } = req.body || {};
    const img = await fetchImage({ image_url, image_base64 });
    const inputHash = hashInput(img.buffer);

    let classification;
    if (use_ai) {
      classification = await classifyWithAI(img.buffer);
    }
    if (!classification) {
      classification = await classifyHeuristic(img.buffer);
    }

    const result = {
      ok: true,
      endpoint: 'classify',
      request_id: requestId,
      category: classification.category,
      confidence: classification.confidence,
      source: classification.source,
      features: classification.features || null,
      inference_time_ms: Date.now() - start,
    };

    await saveResult('classify', requestId, inputHash, result, classification.confidence, { source: img.source, ai_used: use_ai });
    res.json(result);
  } catch (e) {
    console.error('[classify]', e);
    res.status(500).json({ ok: false, error: e.message, request_id: requestId });
  }
});

export default router;