Deep learning-based system development for black pine bast scale detection

Abstract

The prevention of the loss of agricultural resources caused by pests is an important issue. Advances are being made in technologies, but current farm management methods and equipment have not yet met the level required for precise pest control, and most rely on manual management by professional workers. Hence, a pest detection system based on deep learning was developed for the automatic pest density measurement. In the proposed system, an image capture device for pheromone traps was developed to solve nonuniform shooting distance and the reflection of the outer vinyl of the trap while capturing the images. Since the black pine bast scale pest is small, pheromone traps are captured as several subimages and they are used for training the deep learning model. Finally, they are integrated by an image stitching algorithm to form an entire trap image. These processes are managed with the developed smartphone application. The deep learning model detects the pests in the image. The experimental results indicate that the model achieves an F1 score of 0.90 and mAP of 94.7% and suggest that a deep learning model based on object detection can be used for quick and automatic detection of pests attracted to pheromone traps.

Multiclass insect counting through deep learning-based density maps estimation

“…Develop a method for olive fruit fly detection and counting, where insects appear over a controlled yellow background. In the same line [49], propose a method based on performing YOLO over tiles of input images, where black pine bast scale detection is done. There are traps in the image that increase the contrast between the black pest and the yellow background.…”

Insect Detection in Sticky Trap Images of Tomato Crops Using Machine Learning

“…In [68], a pheromone-trapping device was developed. In this work, the original image was cropped into several sub-images with 30% overlap.…”

“…These results seem aligned with the ones achieved in [63]; however, since the images were acquired in a less adverse environment, the results may be biased when compared with those resulting from images acquired directly in the greenhouse. In [68], different object detection models were tested for detecting black pine bast scale pests Among the tested models, YOLOv5 achieved the best results, reaching an F1 score of 0.90 and mAP of 94.7%. The setup used for the image acquisition process (besides being used for a different task) was much more sophisticated than our own.…”

Image-Based Insect Counting Embedded in E-Traps That Learn without Manual Image Annotation and Self-Dispose Captured Insects

“…(A) Deep learning classification largely depends on the availability of a large amount of training examples [29][30][31][32][33][34][35][36][37][38][39]. Construction of large image datasets from real field operation is time-consuming to collect, as it requires annotation (i.e., manually labeling insects with a bounding box using specialized software).…”