This study was conducted to apply a deep learning model to identify and visualise path areas in an orchard. Data was collected by attaching an image capture device to the front of a sprayer and driving it through an orchard. The collected data was classified into four classes: ground, trees, sky and obstacles for pre-processing for training. Sliding window techniques were used on the image dataset for model training and performance. The image was sampled using a sliding window method with 224x224 pixels and divided into train, validation, and test sets. A modified VGG16 algorithm was implemented and used to train the preprocessed image dataset. The performance results of this model showed an accuracy of up to approximately 99% on both the training and validation sets, and after building a confusion matrix using the test set, the classification performance was evaluated and showed an F1 score of 0.96. To visualise the results of this trained learning model, class activation maps were used to detect the paths in the orchard. The implementation of this method resulted in an average processing time of about 0.94 seconds per frame on the orchard footage, which could potentially be beneficial in real-time decision making scenarios in fruit farming where rapid response is important.