Potato Beetle Detection with Real-Time and Deep Learning

In this study, deep learning methods were used to detect potato beetles (Leptinotarsa decemlineata) on potato plants. High-resolution images were taken of fields with the help of a drone. Since these images were large in size, each one was divided into six equal parts. Then, according to the image, the potato beetles were divided into three classes: adult, late-stage potato beetle, and no beetles. A data set was created with 3000 images in each class, making 9000 in total. Different filters were applied to the images that made up the data set. In this way, problems that may have arisen from the camera in real-time detection were minimized. At the same time, the accuracy rate was increased. The created data set was used with six different deep learning models: MobileNet, InceptionV3, ResNet101, AlexNet, DenseNet121, and Xception. The deep learning models were tested with Sgd, Adam, and Rmsprop optimization methods and their performances were compared. In order to evaluate the success of the models more accurately, they were tested on a second data set created with images taken from a different field. As a result of this study, the highest accuracy of 99.81% was obtained. In the test results from a second field that did not exist in the data set, 92.95% accuracy was obtained. The average accuracy rate was 96.30%.