Development of a Specie-specific Bird Deterrent System using Birds Classifications by Convolutional Neural Network (CNN) Model
Abstract
A trained convolutional neural network (CNN) model was developed in this work for classification of birds that visit rice farms into harmful Sparrows or beneficial insectivorous birds, and the classification used in activating efficient pest bird deterrence. Different images of the prevalent pest sparrow were captured by high resolution camera, and used as datasets for the training of the CNN model for the pest bird identification. Since, 98% of sparrow birds are grain eaters and harmful to a rice farm, 2,000 images of different sparrow birds were used together with the 419 different images of the prevalent Sparrows found in the rice farms as datasets for the training of the CNN model in Google Colab platform. A suitable algorithm was developed that uses this birds classification for recognizing the birds that visit the farm as pest Sparrows or otherwise. A bioacoustics deterrent system that uses the recorded sound of a local predator of Sparrows was developed for sparrow-pests. This sound was activated to broadcast the predator sound through the speakers to scare away the birds once Sparrows are sighted in a rice farm or nearby surroundings. However, if the sighted bird is recognized as not a Sparrow, no sound will be activated, so that the beneficial birds will be allowed do their biological insect-pest control in the rice farm. The algorithm can also be used by researchers and teachers in agriculture-related disciplines to teach bird classification in a rice farm.
Keywords
Download Options
Introduction
Two thirds of the Nigerian populace get their livelihood from agriculture; a major source of income and feeding for most developing nations. Presently, it contributes well over a fifth of Nigeria’s gross domestic product (GDP) [1]. Pest infestations in farms seem to be a major part of the numerous problems faced by the agricultural sector [1-8]. Birds are the most destructive pests to in crops farms, and total loss of farm produce can ensue if birds are not controlled [2]. While Java Sparrows are the most destructive bird pests for rice in the farm, there are other birds which feed on insects that attack the rice plants; thus performing biological pest control [2-7]. The different bird deterrent solutions available are not specific in the type of birds to be scared away and some do not deploy the right predator for such birds. With non-selective scaring away of birds, the pestbirds are scared away as well as some of the useful insectivorous birds. These insectivorous birds, by feeding on insects and other rice pests, effect biological control of the pest population; increasing the productivity of rice. Hence a device that can be used for detecting the harmful birds and scare them away while leaving the beneficial ones to remain in the farm will be a welcome development.
Unmanned Aerial Vehicles (UAVs) or drones has be used in capturing, processing, and analyzing images for farm pest management, but have limitations like untimely reach to remote spots, security-related ban, altitude reach limitation and scaring of harmful and useful birds alike [3]. Additional pollution, wind opposition and non-conservation issues arise when chemical deterrence is used with the drones [9]. Drones equipped with multispectral, thermal and visible cameras have been deployed for monitoring in agriculture, but challenges of drone architecture, wind influence on flight, low GPS accuracy have been reported [10].
Conclusion
The developed Sparrow deterrent system used the bird classification into harmful sparrow and beneficial birds by the Convolutional neural network (CNN) model to scare away pest-sparrow bird using predator sound. When the model was used for predictions as shown in Figure 9, it was able to give high accuracy (97.2%), precision (96.5%), recall (0.965) and F1 score (0.965) of the model .Also in a Google Colab platform, the developed model was able to classify birds into harmful sparrow and beneficial (insectivorous) birds. It allowed the beneficial bird in 97% of their visits. Hence, it allowed the beneficial bird to continue its biological pest control without disturbances. However, when the model classification returned harmful Sparrow as shown in Figure 9, it generated the sound of Squirrel which is the predator 97% of the times to scare them away. This was better than the 76% accuracy birds’ classification obtained by a previous researcher [13].