Figures (11)  Tables (8)

    • Figure 1. 

      Work flow of the research methodology.

    • Figure 2. 

      The study area in Garatu Minna, Niger State, Nigeria[28].

    • Figure 3. 

      DJI Mavic drone (Source: https://dronelife.com/2018/01/23/dji-mavic-air, Malek Murison).

    • Figure 4. 

      (a) AlexNet training and validation accuracy. (b) AlexNet training and validation loss.

    • Figure 5. 

      (a) CNN training and validation accuracy. (b) CNN training and validation loss.

    • Figure 6. 

      (a) AlexNet training and validation accuracy (at 40 epochs). (b) AlexNet training and validation loss (at 40 epochs).

    • Figure 7. 

      (a) CNN training and validation accuracy (at 40 epochs). (b) CNN training and validation loss (at 40 epochs).

    • Figure 8. 

      (a) AlexNet training and validation accuracy (at 50 epochs). (b) AlexNet training and validation loss (at 50 epochs).

    • Figure 9. 

      (a) CNN training and validation accuracy (at 50 epochs). (b) CNN training and validation loss (at 50 epochs).

    • Figure 10. 

      (a) AlexNet training and validation accuracy (at 60 epochs). (b) AlexNet training and validation loss (at 60 epochs).

    • Figure 11. 

      (a) CNN training and validation accuracy (at 60 epochs). (b) CNN training and validation loss (at 60 epochs).

    • ParametersSpecificationsRemark
      Number of rotors4Indirectly ensures no omission/gap in captured images
      GSD22.2 mmThis helped to achieve a better image resolution
      Mission time94:17 minThis is the total time taken for the drone to take off, cover the area of interest and return back
      Battery4,000 mAhCapacity of the drone's battery – has a direct impact on cost and time of the flight
      Flight direction−50 °C
      Number of batteries used6The life cycle of a fully charged battery of the drone is less than 16 min, hence, six were used for the flight
      Flying altitude30 mSelected to ensure desired high spatial resolution is achieved
      Flying velocity5 ms−1Velocity set in agreement with the desired image quality
      Flight dateAugust, 2019

      Table 1. 

      Flight specifications.

    • ParametersSpecifications
      ModelMavic 1
      Sensor94:17 min
      Resolution Sensor typeCMOS
      Resolution1.0 cm/px
      F-stopf/2.8
      ISO100
      Focal length3.5 mm

      Table 2. 

      Camera specifications.

    • HyperparametersAlexNetCNN
      Depth8 layers5 layers
      Image size224 × 224224 × 224
      Batch size3232
      No of epochs30−6030−60
      Learning rate0.00010.0001

      Table 3. 

      Hyperparameters for training sample analysis.

    • EpochsTime taken (h)Training accuracy (%)Training lossValidation Accuracy (%)Validation loss
      AlexNetCNNAlexNetCNNAlexNetCNNAlexNetCNNAlexNetCNN
      302.732.0493.7%44.3%2.0432.04561%21%1.2621.682
      404.154.1794.8%54.2%0.0711.40868%39%0.9881.482
      505.585.4599.3%54.5%0.0251.20872%42%1.7451.435
      606.856.6798.6%63%0.0791.02066%47%1.7301.424

      Table 4. 

      AlexNet and CNN training performance.

    • ModelAlexNetCNN
      Training accuracy93.73%44.29%
      Validation accuracy61.07%21.48%
      Validation loss1.26161.6816

      Table 5. 

      AlexNet and CNN behaviour over 30 epochs.

    • ModelAlexNetCNN
      Training accuracy94.84%54.23%
      Validation accuracy67.79%38.93%
      Validation loss0.98791.4820

      Table 6. 

      AlexNet and CNN behaviour at 40 epochs.

    • ModelAlexNetCNN
      Training accuracy99.25%54.53%
      Validation accuracy71.81%42.28%
      Validation loss1.74481.4350

      Table 7. 

      AlexNet and CNN behaviour at 50 epochs.

    • ModelAlexNetCNN
      Training accuracy98.58%62.83%
      Validation accuracy65.77%46.98%
      Validation loss1.73011.4237

      Table 8. 

      AlexNet and CNN behaviour at 60 epochs.