机器学习-CoreML结合Siri

使用AVFoundation拍摄照片，再使用Core ML处理分析照片，最后通过Siri告知拍摄的对象是什么。

1. 前期配置

   var captureView : UIView!
   var synthe = AVSpeechSynthesizer()
       var uttrence = AVSpeechUtterance()
       var predicte = ""
       
       var captureSession : AVCaptureSession!
       var cameraOutput : AVCapturePhotoOutput!
       var previewLayer : AVCaptureVideoPreviewLayer!
       
       override func viewDidLoad() {
           super.viewDidLoad()
           
           captureView = UIView.init(frame: CGRect.init(x: 0, y: 0, width: 200, height: 200))
           self.view.addSubview(captureView)
           setupCamera()
       }
   
       func setupCamera() {
           captureSession = AVCaptureSession()
           captureSession.sessionPreset = AVCaptureSession.Preset.photo
           cameraOutput = AVCapturePhotoOutput()
           
           let device = AVCaptureDevice.default(for: .video)
           if let input = try? AVCaptureDeviceInput.init(device: device!) {
               if(captureSession.canAddInput(input)) {
                   captureSession.addInput(input)
                   
                   if(captureSession.canAddOutput(cameraOutput)) {
                       captureSession.addOutput(cameraOutput)
                   }
                   
                   previewLayer = AVCaptureVideoPreviewLayer.init(session: captureSession)
                   previewLayer.videoGravity = .resizeAspectFill
                   previewLayer.frame = CGRect.init(x: 0, y: 0, width: 200, height: 200)
                   captureView.layer.addSublayer(previewLayer)
                   captureSession.startRunning()
               }
           }
       }

2. 获取照片

   func launchUI() {
       let setting = AVCapturePhotoSettings()
       // xcode9's Bug
       let previewPixelType = setting.__availablePreviewPhotoPixelFormatTypes.first!
       let previewFormat = [ kCVPixelBufferPixelFormatTypeKey as String: previewPixelType,
                             kCVPixelBufferWidthKey as String: "\(captureView.bounds.size.width)",
                             kCVPixelBufferHeightKey as String: "\(captureView.bounds.size.height)"] as [String : Any]
       setting.previewPhotoFormat = previewFormat
       cameraOutput.capturePhoto(with: setting, delegate: self)
   }
   
   // AVCapturePhotoCaptureDelegate
   func photoOutput(_ output: AVCapturePhotoOutput, didFinishProcessingPhoto photo: AVCapturePhoto, error: Error?) {
       if error != nil {
           print("error occured: \(error!.localizedDescription)")
       }
   
       if let imageData = photo.fileDataRepresentation(),let image = UIImage.init(data: imageData) {
          // Predict
          self.predict(image: image)
       }
   }

在Xcode9上availablePreviewPhotoPixelFormatTypes这个属性前面需要加上双下划线,更高版本则不需要

3.图片处理预测

 func predict(image:UIImage) {
//        if let data = UIImagePNGRepresentation(image) {
//            let fileName = getDocumentsDirectory().appendingPathComponent("captureImage")
//            print("fleName:\(fileName)")
//            try? data.write(to: fileName)
//        }
        let model = try! VNCoreMLModel.init(for: VGG16().model)
        let request = VNCoreMLRequest.init(model: model, completionHandler: { (request:VNRequest, error:Error?) in
            weak var weakSelf = self
            guard let results = request.results as? [VNClassificationObservation] else {
                fatalError("no result")
            }
            var bestPrediction = ""
            var bestConfidence:VNConfidence = 0
            
            for classfication:VNClassificationObservation in results {
                if classfication.confidence > bestConfidence {
                    bestConfidence = classfication.confidence
                    bestPrediction = classfication.identifier
                }
            }
          
            weakSelf!.say(string: "\(bestPrediction)")
            
            DispatchQueue.main.asyncAfter(deadline: DispatchTime.now() + 5, execute: {
                weakSelf!.launchUI()
            })
        })
        let handler = VNImageRequestHandler.init(cgImage: image.cgImage!)
        try? handler.perform([request])
    }

4.语音播报

func say(string: String) {
      uttrence = AVSpeechUtterance.init(string: string)
      uttrence.rate = 0.3
      uttrence.voice = AVSpeechSynthesisVoice.init(language: "zh_CN")
      uttrence.pitchMultiplier = 0.8
      uttrence.postUtteranceDelay = 0.2
      synthe.speak(uttrence)
  }

这里面选择的是汉语，不管选择的是什么语音种类，英语都是支持的

5.效果演示

Demo下载

---

如有任何疑问或问题请联系我:fishnewsdream@gmail.com,欢迎交流，共同提高！

Objective-C/Swift技术开发交流群201556264，讨论何种技术并不受限，欢迎各位大牛百家争鸣！

微信公众号OldDriverWeekly，欢迎关注并提出宝贵意见

老司机iOS周报,欢迎关注或订阅

刚刚在线工作室，欢迎关注或提出建设性意见！

刚刚在线论坛, 欢迎踊跃提问或解答！

如有转载，请注明出处，谢谢！