In deeper layers, ResNet heavily depends on skip connections and Relu. Although skip connections have demonstrated their usefulness in networks, a major issue arises when the dimensions between layers are not consistent. In such cases, it is necessary to use techniques such as zero-padding or projection to match the dimensions between layers. These adjustments increase the complexity of the network architecture, resulting in an increase in parameter number and a rise in computational costs. Another problem is the vanishing gradient caused by utilizing Relu. In our model, after making appropriate adjustments to the inception blocks, we replace the deeper layers of ResNet with modified inception blocks and Relu with our non-monotonic activation function (NMAF). To reduce parameter number, we use symmetric factorization and $1\times 1$ convolutions. Utilizing these two techniques contributed to reducing the parameter number by around 6 M parameters, which has helped reduce the run time by 30 s/epoch. Unlike Relu, NMAF addresses the deactivation problem of the non-positive number by activating the negative values and outputting small negative numbers instead of zero in Relu, which helped in enhancing the convergence speed and increasing the accuracy by 5%, 15%, and 5% for the non-noisy datasets, and 5%, 6%, 21% for non-noisy datasets.

Methods to detect directly aphids based on convolutional neural networks (CNNs) are unsatisfactory because aphids are small and usually are specially distributed. To enhance aphid detection efficiency, a framework based on oriented FAST and rotated BRIEF (ORB) and CNNs (EADF) is proposed by us to detect aphids in images. Firstly, the key point is to find regions of aphids. Points generated by the ORB algorithm are processed by us to generate suspected aphid areas. Regions are fed into convolutional networks to train the model. Finally, images are detected in blocks with the trained model. In addition, in order to solve the situation that the coordinates are not uniform after the image is segmented, we use a coordinate mapping method to unify the coordinates. We compare current mainstream target detection methods. Experiments indicate that our method has higher accuracy than state-of-the-art two-stage methods that the AP value of RetinaNet with EADF is 0.385 higher than RetinaNet without it and the Cascade-RCNN with EADF is more than without it by 43.3% on value of AP, which demonstrates its competency.