<p><strong>Deep-learning techniques for symptoms' detection of <em>Aculops lycopersici</em></strong> <strong>(Acari: Eriophyidae) and <em>Tuta absoluta</em></strong> (<strong>Lepidoptera: Gelechiidae) on tomato</strong></p>

Accurate pest detection is the most fundamental requirement in the management of damage-causing factors. Traditional methods of identifying and counting pests in plants require continuous monitoring, which is very costly and time-consuming in large farms, and involves uncontrollable human errors. It is necessary to help farmers identifying key pests automatically and at the beginning phase of the infestation with the help of new technologies. Therefore, deep- learning techniques and a convolutional neural network with VGG Net-16 architecture were used in this study for the automatic detection of the symptoms of two key tomato pests in Iran: Tuta absoluta (Myrick) (Lepidoptera: Gelechiidae) and Aculops lycopersici (Tryon) (Acari: Eriophyidae). A Sony DSC-WX200 camera with an effective sensor resolution of 18 megapixels was used to collect images of the symptoms caused by these pests. To evaluate the performance of the convolutional neural network with VGG Net-16 architecture, the parameters of average precision, precision, and recall were used. To evaluate counting performance, a linear regression curve and the coefficient of determination were used. The detection parameters for the symptoms of T. absoluta and A. lycopersici, including average precision (99.51% and 99.89%, respectively), accuracy (100 for both pests), and recall (100 for both pests), demonstrated the high performance of the convolutional neural network in detecting these two pests. Additionally, the coefficients of determination (0.99 for both pests) indicated the high accuracy of the network in detecting the symptoms of these pests. The results showed that our proposed system can provide a practical solution for the accurate detection of these pests in tomato crops using captured images.